{
  "dataset": "AI training data vendors",
  "description": "Companies that sell training data, RL environments, evaluations, expert work or annotation to AI developers: what each says it sells, the areas it works in and where its data comes from, each with the vendor's own words and the page they are on.",
  "reviewed": "2026-10-15",
  "license": {
    "name": "CC BY 4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/",
    "covers": "our compilation of the tables",
    "excludes": "quotes keep their sources' terms, and article text is not covered"
  },
  "rows": [
    {
      "id": "6thsense",
      "name": "6thSense",
      "website": "https://6thsense.dev/",
      "what": "Builds tactile capture hardware and egocentric cameras for robot learning, and provides datasets recorded with them to partners and customers.",
      "says": {
        "quote": "6thSense builds touch-aware capture hardware for robot learning: Skin, a 464-channel wearable tactile sensor, and Eye2 Pro, an egocentric stereo camera.",
        "url": "https://6thsense.dev/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Provide processed datasets and results to the 6thSense partners and customers those datasets are captured for.",
          "url": "https://6thsense.dev/privacy"
        },
        {
          "kind": "human-data",
          "quote": "6thSense builds tactile-capture hardware and software that records human demonstrations for robotics datasets.",
          "url": "https://6thsense.dev/privacy"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "The capture app records demonstration sessions.",
          "url": "https://6thsense.dev/privacy"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "abundant",
      "name": "Abundant",
      "website": "https://www.abundant.ai/",
      "what": "Creates environments and datasets for reinforcement learning for AI labs and large enterprises.",
      "says": {
        "quote": "We specialize in creating environments and datasets for RL by leveraging our experience in simulation and model training.",
        "url": "https://www.ycombinator.com/companies/abundant"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We specialize in creating environments and datasets for RL by leveraging our experience in simulation and model training.",
          "url": "https://www.ycombinator.com/companies/abundant"
        },
        {
          "kind": "datasets",
          "quote": "Each dataset contains 1000 validated tasks harvested from popular GitHub repos.",
          "url": "https://www.abundant.ai/blog/swe-gen"
        },
        {
          "kind": "evals",
          "quote": "We build the evals, the environments, and the systems that make hill climbing possible at scale.",
          "url": "https://www.abundant.ai/blog/hillclimbing-to-an-abundant-future"
        },
        {
          "kind": "human-data",
          "quote": "Exclusive, global network of over 500 domain experts",
          "url": "https://www.ycombinator.com/companies/abundant"
        }
      ],
      "areas": [
        "coding",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Exclusive, global network of over 500 domain experts",
          "url": "https://www.ycombinator.com/companies/abundant"
        },
        {
          "kind": "public",
          "quote": "Each dataset contains 1000 validated tasks harvested from popular GitHub repos.",
          "url": "https://www.abundant.ai/blog/swe-gen"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, CA",
        "quote": "Based in San Francisco, CA.",
        "url": "https://www.ycombinator.com/companies/abundant"
      },
      "founded": {
        "value": 2024,
        "quote": "Founded in 2024 by Meji Abidoye, Jesse Hu, and Ke Huang, Abundant has 7 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/abundant"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "afterquery",
      "name": "AfterQuery",
      "website": "https://afterquery.com",
      "what": "Sells expert-built training data, RL environments, and custom evaluations to frontier AI labs and enterprises.",
      "says": {
        "quote": "AfterQuery is an applied research lab curating data solutions for frontier foundation model development.",
        "url": "https://www.ycombinator.com/companies/afterquery"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "enabling training and evaluation of agents in real workflows.",
          "url": "https://afterquery.com"
        },
        {
          "kind": "datasets",
          "quote": "Offers ready-to-deploy datasets across high-demand capability areas",
          "url": "https://www.afterquery.com/products"
        },
        {
          "kind": "human-data",
          "quote": "Captured from domain experts step by step, decision by decision, and verified before delivery.",
          "url": "https://www.afterquery.com/solutions"
        },
        {
          "kind": "evals",
          "quote": "We create evaluation task sets that pinpoint exactly where a model breaks down, across any domain or workflow.",
          "url": "https://www.afterquery.com/solutions"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We work with domain experts to capture that thinking",
          "url": "https://afterquery.com"
        },
        {
          "kind": "captured",
          "quote": "Captured from domain experts step by step, decision by decision, and verified before delivery.",
          "url": "https://www.afterquery.com/solutions"
        },
        {
          "kind": "synthetic",
          "quote": "We curate expert-generated datasets, generate synthetic data, and help turn messy enterprise data into a form models can be trained on.",
          "url": "https://www.afterquery.com/solutions"
        },
        {
          "kind": "customer",
          "quote": "help turn messy enterprise data into a form models can be trained on",
          "url": "https://www.afterquery.com/solutions"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, CA",
        "quote": "We are based in San Francisco, CA, and have raised funding from top investors",
        "url": "https://www.ycombinator.com/companies/afterquery/jobs/iTt13cJ-strategic-projects-lead"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Carlos Georgescu and Spencer Mateega, AfterQuery has 30 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/afterquery"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "align-ai",
      "name": "Align AI",
      "website": "https://alignai.co",
      "what": "Builds verification layers, RL environments and expert-sourced ground truth for AI agents; works with frontier labs and enterprises deploying agents.",
      "says": {
        "quote": "ground truth from encoded law, enterprise records, and expert judgment.",
        "url": "https://alignai.co"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "with reward signals grounded in encoded regulation, systems of record, and credentialed expert judgment.",
          "url": "https://alignai.co"
        },
        {
          "kind": "evals",
          "quote": "We compose these sources into verification layers and evaluation frameworks",
          "url": "https://alignai.co"
        },
        {
          "kind": "human-data",
          "quote": "Where ground truth exists only in human judgment, we source, verify, and manage hundreds of licensed domain experts",
          "url": "https://alignai.co"
        }
      ],
      "areas": [
        "enterprise",
        "legal",
        "medical",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We extract ground truth from what credentialed experts do, not just what they say",
          "url": "https://alignai.co"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "andon-labs",
      "name": "Andon Labs",
      "website": "https://andonlabs.com",
      "what": "Develops custom evaluations for AI models and runs benchmarks such as Vending-Bench and Drone-Bench.",
      "says": {
        "quote": "Andon Labs develops custom evaluations for AI models",
        "url": "https://andonlabs.com/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Andon Labs develops custom evaluations for AI models",
          "url": "https://andonlabs.com/"
        }
      ],
      "areas": [
        "computer-use",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "The data comes from our deployment of real-world autonomous organizations such as Andon Market, Andon Café, and Andon FM",
          "url": "https://andonlabs.com/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "andromede",
      "name": "Andromede",
      "website": "https://andromede.ai",
      "what": "Builds RL environments, tasks and verifiers from real-world data for post-training and evaluation of frontier agents.",
      "says": {
        "quote": "Andromede is an RL data lab generating environments, tasks, and verifiers from real-world data.",
        "url": "https://andromede.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Programmatic generation of RL environments for post-training and evaluation.",
          "url": "https://andromede.ai"
        },
        {
          "kind": "evals",
          "quote": "Programmatic generation of RL environments for post-training and evaluation.",
          "url": "https://andromede.ai"
        }
      ],
      "areas": [],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "anthromind",
      "name": "Anthromind",
      "website": "https://anthromind.com",
      "what": "Provides expert evaluations, RLHF preference data and post-training datasets for teams building and evaluating AI models.",
      "says": {
        "quote": "High-quality data needed to evaluate, supervise and align AI systems.",
        "url": "https://anthromind.com"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Build tailored evaluation processes to confidently evaluate your LLM workflows, backed by research, comparative, custom, or industry-standard benchmarks.",
          "url": "https://www.anthromind.com/use-cases"
        },
        {
          "kind": "datasets",
          "quote": "Create custom high quality, premium datasets to enhance AI model performance by improving accuracy, domain expertise or to provide a consistent tone or style.",
          "url": "https://www.anthromind.com/use-cases"
        },
        {
          "kind": "human-data",
          "quote": "Refine model outputs using expert feedback integrated for reinforcement learning.",
          "url": "https://www.anthromind.com/foundation-models"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "legal"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Get responses for hard problems across domains (coding, STEM, agentic use cases, etc) curated by experts.",
          "url": "https://www.anthromind.com/foundation-models"
        },
        {
          "kind": "customer",
          "quote": "Leverage Anthromind experts to convert your data into business value assuring your competitive edge.",
          "url": "https://anthromind.com/enterprise-ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "appen",
      "name": "Appen",
      "website": "https://appen.com",
      "what": "Provides AI training data, data annotation and model evaluation to AI labs and other AI teams.",
      "says": {
        "quote": "We deliver the expert-validated data that trains frontier models, ensuring AI systems understand nuance, context, and complexity at scale.",
        "url": "https://appen.com"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Appen provides specialist LLM training data across the full development pipeline, from initial fine-tuning demonstrations through to the expert preference feedback and adversarial testing that alignment requires.",
          "url": "https://www.appen.com/llm-training-data"
        },
        {
          "kind": "human-data",
          "quote": "Expert preference rankings, comparative evaluations, and nuanced feedback from verified subject matter experts across medicine, law, science, and finance for reinforcement learning with human feedback.",
          "url": "https://www.appen.com/llm-training-data"
        },
        {
          "kind": "rl-environments",
          "quote": "Appen's RL environment design service provides end-to-end environment construction for teams training agents on verifiable tasks where automated reward computation is feasible.",
          "url": "https://www.appen.com/reinforcement-learning-environments"
        },
        {
          "kind": "evals",
          "quote": "Hallucination benchmarking, regulatory audits, bias detection and continuous monitoring to ensure your models are trusted.",
          "url": "https://appen.com"
        },
        {
          "kind": "annotation",
          "quote": "LiDAR point cloud annotation, multi-camera sensor fusion, robot demonstration trajectories, world model rollouts and embodied interaction logs for AI systems operating in unstructured physical environments.",
          "url": "https://appen.com"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "robotics",
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Expert preference rankings, comparative evaluations, and nuanced feedback from verified subject matter experts across medicine, law, science, and finance for reinforcement learning with human feedback.",
          "url": "https://www.appen.com/llm-training-data"
        },
        {
          "kind": "licensed",
          "quote": "Enterprise sourced data from a wide-range of enterprises, covering different environments, industries and time-horizons",
          "url": "https://www.appen.com/enterprise-data"
        },
        {
          "kind": "synthetic",
          "quote": "Context engineering, calibrated LLM-as-a-judge deployment, synthetic data generation and custom benchmarking for teams extending their research capacity.",
          "url": "https://appen.com"
        }
      ],
      "buysData": {
        "quote": "Terms are agreed in writing before extraction begins, and payment is on acceptance rather than submission.",
        "url": "https://www.appen.com/enterprise-data"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-16"
    },
    {
      "id": "applied-intuition",
      "name": "Applied Intuition",
      "website": "https://appliedintuition.com",
      "what": "Builds physical AI software, including simulation and autonomy tools, for machines in automotive, defense, trucking, mining, construction and agriculture.",
      "says": {
        "quote": "Applied Intuition powers physical AI, automating machines across automotive, defense, trucking, mining, construction, & agriculture.",
        "url": "https://appliedintuition.com"
      },
      "sells": [
        {
          "kind": "simulation",
          "quote": "Axion - Sim: High-fidelity effects-based and physics-based simulation with features across all domains",
          "url": "https://www.appliedintuition.com/blog/axion-and-acuity"
        },
        {
          "kind": "rl-environments",
          "quote": "Axion - RL: A deep reinforcement learning league play tool for training AI agents",
          "url": "https://www.appliedintuition.com/blog/axion-and-acuity"
        },
        {
          "kind": "evals",
          "quote": "Evaluation of multi-domain collaborative autonomy in Axion Sim",
          "url": "https://www.appliedintuition.com/blog/axion-and-acuity"
        },
        {
          "kind": "annotation",
          "quote": "Drive autonomy stack performance with a closed-loop system that converts real-world sensor data into curated, labeled segments for downstream training, simulation, and evaluation tasks.",
          "url": "https://www.appliedintuition.com/products"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Our end-to-end physical AI platform uses simulation and real-world data collected across industries",
          "url": "https://www.appliedintuition.com/products"
        },
        {
          "kind": "synthetic",
          "quote": "Synthetic data enables the scalable generation and verification of situations by changing parameters such as weather, fog, time of day, and visibility.",
          "url": "https://www.appliedintuition.com/case-studies/toyota"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "aptura",
      "name": "Aptura",
      "website": "https://aptura.ai",
      "what": "Helps AI researchers train and evaluate models with datasets, RL environments and domain experts for finance, healthcare and legal tasks.",
      "says": {
        "quote": "We help train and evaluate AI to master real-world tasks in finance, healthcare, and legal.",
        "url": "https://aptura.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "verifiable rewards, custom tooling, data licensing support.",
          "url": "https://aptura.ai"
        },
        {
          "kind": "datasets",
          "quote": "High-quality datasets. Domain-specific data that reflects the real complexity of regulated work.",
          "url": "https://aptura.ai"
        },
        {
          "kind": "human-data",
          "quote": "Physicians, attorneys, and investment analysts with years of high-stakes experience, supported by a platform built for structured, high-quality data collection at scale.",
          "url": "https://aptura.ai"
        },
        {
          "kind": "evals",
          "quote": "We help train and evaluate AI to master real-world tasks in finance, healthcare, and legal.",
          "url": "https://aptura.ai"
        }
      ],
      "areas": [
        "finance",
        "legal",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Physicians, attorneys, and investment analysts with years of high-stakes experience, supported by a platform built for structured, high-quality data collection at scale.",
          "url": "https://aptura.ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "archal",
      "name": "Archal",
      "website": "https://archal.ai/",
      "what": "Provides stateful API sandboxes with environments for services like GitHub and Slack, for AI agent developers and integration teams to test against.",
      "says": {
        "quote": "Each sandbox contains stateful environments that behave like the APIs you use in production.",
        "url": "https://archal.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Give every rollout a sandbox with the same starting state while changes persist through the episode.",
          "url": "https://archal.ai/"
        },
        {
          "kind": "evals",
          "quote": "Create a sandbox with the API environments your CI, tests, and evals need.",
          "url": "https://archal.ai/product"
        }
      ],
      "areas": [
        "coding"
      ],
      "madeFrom": [
        {
          "kind": "replica",
          "quote": "Each environment is an independently built simulation of the service it represents, not the provider’s live service.",
          "url": "https://archal.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Aidan Tiruvan, Archal has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/archal"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "arena",
      "name": "Arena",
      "website": "https://arena.ai/",
      "what": "Sells AI model evaluations based on human feedback to enterprises, model labs and developers.",
      "says": {
        "quote": "This service offers enterprises, model labs, and developers comprehensive evaluation services grounded in real-world human feedback, showing how models actually perform in practice.",
        "url": "https://arena.ai/blog/ai-evaluations"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "This service offers enterprises, model labs, and developers comprehensive evaluation services grounded in real-world human feedback, showing how models actually perform in practice.",
          "url": "https://arena.ai/blog/ai-evaluations"
        },
        {
          "kind": "datasets",
          "quote": "Arena operates the largest human-preference dataset for AI evaluation in the world.",
          "url": "https://arena.ai/blog/arena-100m-revenue"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "public",
          "quote": "Key datasets and foundational research will remain publicly available.",
          "url": "https://arena.ai/blog/ai-evaluations"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "arga-labs",
      "name": "Arga Labs",
      "website": "https://www.argalabs.com",
      "what": "Sells stateful sandbox twins of third-party APIs, CLIs and MCPs, used by teams to train and test AI agents.",
      "says": {
        "quote": "Deploy simulation environments with stateful twins of the APIs, CLIs, and MCPs your agents use.",
        "url": "https://www.argalabs.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Arga gives models multi-app workflows where they can practice real tool use, acquire skills, and improve before touching production.",
          "url": "https://www.argalabs.com/use-cases"
        },
        {
          "kind": "evals",
          "quote": "Automate evals after agents run in Arga. Define judging criteria and grade performance for every commit or PR.",
          "url": "https://www.argalabs.com"
        }
      ],
      "areas": [
        "enterprise",
        "coding"
      ],
      "madeFrom": [
        {
          "kind": "replica",
          "quote": "Full clones with API, MCP, and CLI support for third-party services your agents need to read from and write to.",
          "url": "https://www.argalabs.com"
        },
        {
          "kind": "customer",
          "quote": "Describe the scenario in natural language and get a seeded service with mock data, or seed the services with real data from prod.",
          "url": "https://www.argalabs.com"
        },
        {
          "kind": "synthetic",
          "quote": "describe what you want and Arga generates the seed configuration automatically",
          "url": "https://docs.argalabs.com/features/custom-scenarios.md"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "arimlabs",
      "name": "ARIMLABS",
      "website": "https://arimlabs.ai",
      "what": "Builds RLVR training environments and benchmarks that measure what frontier AI models and agents actually do on long-horizon tasks.",
      "says": {
        "quote": "ARIMLABS builds production-fidelity simulations where AI agents run multi-hour tasks, and the benchmarks that measure what frontier models actually do.",
        "url": "https://arimlabs.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Carefully calibrated training environments for the agents of tomorrow.",
          "url": "https://arimlabs.ai"
        },
        {
          "kind": "evals",
          "quote": "We measure what frontier models can actually achieve on the bleeding edge of long-horizon, domain-specific tasks across cybersecurity, SRE, terminal use, STEM, and more.",
          "url": "https://arimlabs.ai"
        },
        {
          "kind": "simulation",
          "quote": "ARIMLABS builds production-fidelity simulations where AI agents run multi-hour tasks, and the benchmarks that measure what frontier models actually do.",
          "url": "https://arimlabs.ai"
        }
      ],
      "areas": [
        "security",
        "coding"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "asimov",
      "name": "Asimov",
      "website": "https://tryasimov.ai/",
      "what": "Collects real-world human activity recordings and turns them into annotated datasets for training robots, sold to robotics teams.",
      "says": {
        "quote": "We collect diverse real-world human data at scale to power the next generation of robots.",
        "url": "https://tryasimov.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We collect human activity data at scale from real environments across multiple continents, and turn it into clean, annotated datasets that teach robots to understand the physical world.",
          "url": "https://tryasimov.ai/"
        },
        {
          "kind": "human-data",
          "quote": "We collect diverse real-world human data at scale to power the next generation of robots.",
          "url": "https://tryasimov.ai/"
        },
        {
          "kind": "annotation",
          "quote": "proprietary hardware, annotation tooling, and a vertically integrated pipeline",
          "url": "https://www.ycombinator.com/companies/asimov"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "First-person RGB video recorded during assigned tasks.",
          "url": "https://tryasimov.ai/privacy"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, US",
        "quote": "Founded in 2026 by Lyem Ningthou, Asimov has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/asimov"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Lyem Ningthou, Asimov has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/asimov"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "aviro",
      "name": "Aviro",
      "website": "https://www.aviro.ai",
      "what": "Builds post-training datasets and reinforcement learning environments for frontier AI labs.",
      "says": {
        "quote": "We turn these worlds into post-training datasets and reinforcement learning environments to train the next frontier of AI.",
        "url": "https://www.aviro.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We turn these worlds into post-training datasets and reinforcement learning environments to train the next frontier of AI.",
          "url": "https://www.aviro.ai/"
        },
        {
          "kind": "datasets",
          "quote": "We turn these worlds into post-training datasets and reinforcement learning environments to train the next frontier of AI.",
          "url": "https://www.aviro.ai/"
        },
        {
          "kind": "evals",
          "quote": "Aviro benchmarks measure how well frontier models handle document-heavy workflows, grounded reasoning, and multi-step tool use inside enterprise environments.",
          "url": "https://www.aviro.ai/benchmarks"
        },
        {
          "kind": "simulation",
          "quote": "three simulated enterprise environments: a SaaS analytics platform, a financial services firm, and a chemical manufacturing company",
          "url": "https://www.aviro.ai/writeups/c4-ebla"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Domain specialists authored all tasks and verified every rubric criterion.",
          "url": "https://www.aviro.ai/writeups/c4-ebla"
        },
        {
          "kind": "replica",
          "quote": "Platform engineers build replicas of platforms, populated by data from the knowledge graph.",
          "url": "https://www.aviro.ai/writeups/c4-ebla"
        },
        {
          "kind": "synthetic",
          "quote": "Each is generated from a knowledge graph that serves as the single source of truth.",
          "url": "https://www.aviro.ai/writeups/c4-ebla"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2024 by Ayush Karupakula and Aarush Kukreja, Aviro has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/aviro"
      },
      "founded": {
        "value": 2024,
        "quote": "Founded in 2024 by Ayush Karupakula and Aarush Kukreja, Aviro has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/aviro"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "benchflow",
      "name": "BenchFlow",
      "website": "https://benchflow.ai",
      "what": "Builds environments and benchmarks that AI agents learn and are evaluated in; works with frontier labs.",
      "says": {
        "quote": "We build the environments AI agents need to learn and be evaluated on real computer work, not static prompts.",
        "url": "https://www.benchflow.ai/about"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "The first benchmark for evaluating AI agent skills. 87 tasks, 24 model-harness configurations, 3 trials per task.",
          "url": "https://www.benchflow.ai/research/skillsbench"
        },
        {
          "kind": "simulation",
          "quote": "Coding agents drive simulated robots from the shell.",
          "url": "https://benchflow.ai"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "robotics",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "100+ expert-curated tasks, built with Kaggle.",
          "url": "https://benchflow.ai"
        },
        {
          "kind": "replica",
          "quote": "five high-fidelity mock services that replicate real Google Workspace and Slack APIs",
          "url": "https://www.benchflow.ai/research/clawsbench"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2024,
        "quote": "Started in late 2024. SkillsBench has 1,800+ GitHub stars and 260+ citations on Semantic Scholar",
        "url": "https://www.benchflow.ai/about"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "besimple-ai",
      "name": "Besimple AI",
      "website": "https://besimple.ai",
      "what": "Supplies licensed conversational audio datasets, annotation, and evaluation support to voice AI teams building speech models.",
      "says": {
        "quote": "Besimple helps voice AI teams collect, license, annotate, and evaluate human audio data without building an internal data operation from scratch.",
        "url": "https://besimple.ai/audio-data"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We source long-form, natural conversations between real speakers so voice models learn interruptions, turn-taking, accent variation, and messy production speech.",
          "url": "https://besimple.ai/conversational-audio-data"
        },
        {
          "kind": "human-data",
          "quote": "Besimple helps voice AI teams collect, license, annotate, and evaluate human audio data without building an internal data operation from scratch.",
          "url": "https://besimple.ai/audio-data"
        },
        {
          "kind": "annotation",
          "quote": "Add transcription, diarization, quality review, emotion tags, speaker metadata, and domain labels so training data is usable immediately.",
          "url": "https://besimple.ai/audio-data"
        },
        {
          "kind": "evals",
          "quote": "A public, test-only benchmark for evaluating whether AI audio models can identify expressed vocal emotion from raw speech",
          "url": "https://besimple.ai"
        },
        {
          "kind": "rl-environments",
          "quote": "Environments built using real human speech to improve instruction following and conversational handling.",
          "url": "https://besimple.ai"
        }
      ],
      "areas": [
        "speech",
        "multilingual",
        "finance",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "We source long-form, natural conversations between real speakers so voice models learn interruptions, turn-taking, accent variation, and messy production speech.",
          "url": "https://besimple.ai/conversational-audio-data"
        },
        {
          "kind": "experts",
          "quote": "We then leverage human expert audio annotators and our own annotation platform to process audio data for Automatic Speech Recognition.",
          "url": "https://www.ycombinator.com/companies/besimple-ai"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, United States",
        "quote": "Founded in 2025 by Bill Wang and Yi Zhong, Besimple AI has 10 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/besimple-ai"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Bill Wang and Yi Zhong, Besimple AI has 10 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/besimple-ai"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "bespoke-labs",
      "name": "Bespoke Labs",
      "website": "https://bespokelabs.ai",
      "what": "Builds reinforcement learning environments and infrastructure for frontier labs and enterprises to train long-horizon agents.",
      "says": {
        "quote": "Bespoke Labs builds RL environments and infrastructure to let frontier labs and enterprises train long-horizon agents for production.",
        "url": "https://bespokelabs.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Bespoke Labs builds RL environments and infrastructure to let frontier labs and enterprises train long-horizon agents for production.",
          "url": "https://bespokelabs.ai/"
        },
        {
          "kind": "evals",
          "quote": "Evaluate agents in environments that mirror your systems and processes.",
          "url": "https://bespokelabs.ai/"
        },
        {
          "kind": "datasets",
          "quote": "One of the best open reasoning datasets.",
          "url": "https://bespokelabs.ai/"
        },
        {
          "kind": "simulation",
          "quote": "Company-scale simulations with real codebases and microservices, allowing agents to master the complex, long-horizon workflows required for production.",
          "url": "https://bespokelabs.ai/"
        }
      ],
      "areas": [
        "coding",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Domain experts interested in building RL environments and data pipelines.",
          "url": "https://bespokelabs.ai/"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Mountain View, United States",
        "quote": "800 W El Camino Real, Suite 225, Mountain View, CA 94040",
        "url": "https://bespokelabs.ai/about"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "biostack-platforms",
      "name": "BioStack Platforms",
      "website": "https://www.getbiostack.com/",
      "what": "Sells healthcare datasets, RL environments, and evaluations to AI labs for training and testing models.",
      "says": {
        "quote": "Get novel pre-clinical and medical datasets, perform causal inference, and generate data points for pre / post training.",
        "url": "https://www.getbiostack.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Build RL envs for post-training",
          "url": "https://www.getbiostack.com/"
        },
        {
          "kind": "datasets",
          "quote": "Get novel pre-clinical and medical datasets, perform causal inference, and generate data points for pre / post training.",
          "url": "https://www.getbiostack.com/"
        },
        {
          "kind": "evals",
          "quote": "Data, evals, reward functions, benchmarks, and simulation environments, so models can practice real clinical work instead of just memorising medical exams.",
          "url": "http://ycombinator.com/companies/biostack-platforms"
        },
        {
          "kind": "simulation",
          "quote": "Our environments use real-world clinical data to help models practice solving the hardest medical problems.",
          "url": "http://ycombinator.com/companies/biostack-platforms"
        }
      ],
      "areas": [
        "medical",
        "science",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "We started with our proprietary dataset of 186 de-identified CT, MRI, and PET/CT cases paired with radiologist reports.",
          "url": "https://www.getbiostack.com/blog/we-made-medical-ai-show-its-work-part1"
        },
        {
          "kind": "experts",
          "quote": "Two multimodal reviewers scanned the sheets independently without seeing the target finding and nominated candidate slices.",
          "url": "https://www.getbiostack.com/blog/we-made-medical-ai-show-its-work-part1"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "BioStack Platforms has 10 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/biostack-platforms"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Sanat Mishra and Parth Patwa, BioStack Platforms has 10 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/biostack-platforms"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "bugcrowd",
      "name": "Bugcrowd",
      "website": "https://bugcrowd.com",
      "what": "Runs crowdsourced bug bounties and penetration tests for enterprises, and sells RL environments to AI labs.",
      "says": {
        "quote": "Find and fix hidden vulnerabilities faster by accessing the world’s best hackers and pentesters.",
        "url": "https://www.bugcrowd.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "The platform includes hundreds of thousands of training environments, each built from authentic open-source vulnerabilities with real source code and verifiable outcomes",
          "url": "https://www.bugcrowd.com/press-release/bugcrowd-launches-reinforcement-learning-environments-to-help-ai-models-learn-real-world-security-skills"
        }
      ],
      "areas": [
        "security",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "public",
          "quote": "All environments are derived exclusively from open-source software, and no customer data or security researchers are used at any stage of the training process.",
          "url": "https://www.bugcrowd.com/press-release/bugcrowd-launches-reinforcement-learning-environments-to-help-ai-models-learn-real-world-security-skills"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, United States",
        "quote": "Bugcrowd, the leader in preemptive cybersecurity, today announced",
        "url": "https://www.bugcrowd.com/press-release/bugcrowd-launches-reinforcement-learning-environments-to-help-ai-models-learn-real-world-security-skills"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "build-ai",
      "name": "Build AI",
      "website": "https://www.build.ai/",
      "what": "Collects egocentric factory-worker video and offers it as training datasets for robotics and physical AI researchers and labs.",
      "says": {
        "quote": "Egocentric-10K is the largest egocentric dataset. It is the first dataset collected exclusively in real factories.",
        "url": "https://huggingface.co/datasets/builddotai/Egocentric-10K"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Egocentric-10K is the largest egocentric dataset. It is the first dataset collected exclusively in real factories.",
          "url": "https://huggingface.co/datasets/builddotai/Egocentric-10K"
        },
        {
          "kind": "evals",
          "quote": "The complete 30,000 frame evaluation set is available at Egocentric-10K-Evaluation.",
          "url": "https://huggingface.co/datasets/builddotai/Egocentric-10K"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "centific",
      "name": "Centific",
      "website": "https://centific.com",
      "what": "Provides training data, RL environments, human feedback and evaluations to model labs and enterprises building AI systems.",
      "says": {
        "quote": "We help model labs and enterprises build, train, deploy, and govern intelligent systems through high-quality data, human expertise, and end-to-end platforms that turn complexity into scalable, real-world impact.",
        "url": "https://centific.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Centific designs and operates high-fidelity reinforcement learning environments with human-in-the-loop agents that mirror real-world complexity.",
          "url": "https://centific.com"
        },
        {
          "kind": "datasets",
          "quote": "Centific’s Data Marketplace provides a centralized, self-service platform to discover, enhance, and deliver trusted datasets from both Centific and third-party sources at scale.",
          "url": "https://www.centific.com/products/ai-data-marketplace"
        },
        {
          "kind": "human-data",
          "quote": "We power RLHF pipelines at scale by combining expert raters, multilingual communities, safety frameworks, and proprietary orchestration.",
          "url": "https://centific.com"
        },
        {
          "kind": "evals",
          "quote": "Centific delivers large-scale, statistically valid human evaluation across quality, safety, bias, relevance, and task success.",
          "url": "https://centific.com"
        },
        {
          "kind": "annotation",
          "quote": "Through built-in human-in-the-loop workflows, teams can request labeling, annotation, filtering, and customization on any dataset",
          "url": "https://www.centific.com/products/ai-data-marketplace"
        }
      ],
      "areas": [
        "enterprise",
        "finance",
        "medical",
        "multilingual",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "We license enterprise data across industries and business functions.",
          "url": "https://www.centific.com/enterprise-data"
        },
        {
          "kind": "customer",
          "quote": "whether sourced from the marketplace or uploaded from their own systems, ensuring data is model-ready before it enters training pipelines.",
          "url": "https://www.centific.com/products/ai-data-marketplace"
        },
        {
          "kind": "experts",
          "quote": "We power RLHF pipelines at scale by combining expert raters, multilingual communities, safety frameworks, and proprietary orchestration.",
          "url": "https://centific.com"
        }
      ],
      "buysData": {
        "quote": "We want to partner with enterprises to anonymize and license your operational data for AI training by frontier labs.",
        "url": "https://www.centific.com/enterprise-data"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "chakra-labs",
      "name": "Chakra Labs",
      "website": "https://www.chakra.dev/",
      "what": "Builds expert-authored tasks, benchmarks and training datasets for companies training frontier AI models.",
      "says": {
        "quote": "We build task factories for training frontier models and benchmarks for evaluating them.",
        "url": "https://www.chakra.dev/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Each domain runs as its own factory: tasks written by experts, graders built for real work.",
          "url": "https://www.chakra.dev/"
        },
        {
          "kind": "human-data",
          "quote": "Legal tasks authored by seasoned practicing counsel in a wide range of practice areas.",
          "url": "https://www.chakra.dev/"
        },
        {
          "kind": "evals",
          "quote": "Chakra Labs has built a new benchmark to measure this capability directly: MAGI (MCP and GUI Benchmark).",
          "url": "https://www.chakra.dev/magi-bench"
        },
        {
          "kind": "datasets",
          "quote": "This dataset contains authentic computer interaction data collected from users performing real work tasks in productivity applications.",
          "url": "https://huggingface.co/datasets/chakra-labs/pango-office-trajectories"
        }
      ],
      "areas": [
        "legal",
        "finance",
        "enterprise",
        "computer-use"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Tasks in MAGI are written by expert operators who spend two to three days immersed in the full Company Soul environment before writing a single prompt.",
          "url": "https://www.chakra.dev/magi-bench"
        },
        {
          "kind": "replica",
          "quote": "MAGI consists of 1,000 (200 test / 800 train) tasks mimicking real work spread across 8 cloned enterprise applications.",
          "url": "https://www.chakra.dev/magi-bench"
        },
        {
          "kind": "captured",
          "quote": "The data was collected through Pango, a crowdsourced platform where users are compensated for contributing their natural computer interactions during actual work sessions.",
          "url": "https://huggingface.co/datasets/chakra-labs/pango-office-trajectories"
        },
        {
          "kind": "synthetic",
          "quote": "Synthetically generated task instruction for training purposes. Represents the context of the full task.",
          "url": "https://huggingface.co/datasets/chakra-labs/pango-office-trajectories"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "claru",
      "name": "Claru",
      "website": "https://claru.ai/",
      "what": "Reka AI's training-data business: sells training data, including egocentric video and annotations, to robotics and AI labs.",
      "says": {
        "quote": "Purpose-built datasets for frontier robotics, embodied AI, and world models.",
        "url": "https://claru.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Explore 100+ licensed datasets across robotics, video, and multimodal AI.",
          "url": "https://claru.ai/about"
        },
        {
          "kind": "human-data",
          "quote": "Expert humans label what machines miss: intent, context, edge cases.",
          "url": "https://claru.ai/"
        },
        {
          "kind": "annotation",
          "quote": "2M+ video annotations powering RLHF for frontier video generation.",
          "url": "https://claru.ai/"
        },
        {
          "kind": "evals",
          "quote": "39,000 pairwise human evaluations across 51 model configurations",
          "url": "https://claru.ai/case-studies"
        }
      ],
      "areas": [
        "robotics",
        "vision",
        "games"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "not synthetic, not scraped.",
          "url": "https://claru.ai/"
        },
        {
          "kind": "captured",
          "quote": "A global network of trained data collectors capturing real-world video across 6 continents.",
          "url": "https://claru.ai/"
        },
        {
          "kind": "experts",
          "quote": "Expert humans label what machines miss: intent, context, edge cases.",
          "url": "https://claru.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "cloudfactory",
      "name": "CloudFactory",
      "website": "https://cloudfactory.com",
      "what": "Provides AI consulting, data labeling and annotation, and AI oversight services for enterprises building AI.",
      "says": {
        "quote": "CloudFactory helps enterprises control, validate and scale AI in high stakes environments.",
        "url": "https://cloudfactory.com"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We transform messy, unstructured, or incomplete data into high-quality datasets, creating the foundation for reliable AI models.",
          "url": "https://cloudfactory.com"
        },
        {
          "kind": "annotation",
          "quote": "CloudFactory data analysts precisely annotate medical imaging data to give you reliable object recognition for safe medical devices.",
          "url": "https://www.cloudfactory.com/medical-devices"
        },
        {
          "kind": "human-data",
          "quote": "We screen, hire, train, and manage data analysts to maintain performance and scalability as your data volumes change.",
          "url": "https://www.cloudfactory.com/medical-devices"
        },
        {
          "kind": "evals",
          "quote": "We reduce AI risks with oversight, validation, and error handling combined with human expertise to ensure accuracy and optimize AI systems where failure isn’t an option.",
          "url": "https://cloudfactory.com"
        }
      ],
      "areas": [
        "enterprise",
        "medical",
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "A fully managed workforce experienced in using client-developed and all major 2-D and 3-D tools.",
          "url": "https://www.cloudfactory.com/medical-devices"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "collinear",
      "name": "Collinear",
      "website": "https://collinear.ai",
      "what": "Sells RL environments, verifiers, and training data to frontier AI labs for model improvement.",
      "says": {
        "quote": "Collinear AI builds the realistic RL environments, verifiers, and training data that the largest frontier labs trust for their hardest model improvement problems.",
        "url": "http://collinear.ai/about-us"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Collinear AI builds realistic RL environments, verifiers, and training data for frontier models.",
          "url": "http://collinear.ai/about-us"
        },
        {
          "kind": "datasets",
          "quote": "Training data and reward signals to post-train your model or improve your agent harness",
          "url": "https://www.collinear.ai/"
        },
        {
          "kind": "evals",
          "quote": "Tests coding agents on real vulnerabilities in real codebases.",
          "url": "https://www.collinear.ai/"
        },
        {
          "kind": "simulation",
          "quote": "Every simulation lab is a self-contained world where your agent operates, complete with the users, tools, data, and tasks it will face in production.",
          "url": "https://www.collinear.ai/"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "replica",
          "quote": "Stateful sandboxes / clones of enterprise tools with state persistence across turns.",
          "url": "https://www.collinear.ai/"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Sunnyvale, United States",
        "quote": "Trusted by leading frontier labs, Collinear AI is headquartered in Sunnyvale, California, and is opening a San Francisco office.",
        "url": "http://collinear.ai/about-us"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "contra-labs",
      "name": "Contra Labs",
      "website": "https://contralabs.com",
      "what": "Provides human creative-expert evaluations, benchmarks and training data for creative AI, sold to AI labs and tools.",
      "says": {
        "quote": "A frontier human data and evaluation lab making AI better for creativity.",
        "url": "https://contralabs.com"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "The industry standard for measuring creative AI.",
          "url": "https://contralabs.com"
        },
        {
          "kind": "datasets",
          "quote": "Seed datasets (image, video, UI, motion). Corrective SFT data and RLHF preference pairs.",
          "url": "https://contralabs.com"
        },
        {
          "kind": "human-data",
          "quote": "Seed datasets (image, video, UI, motion). Corrective SFT data and RLHF preference pairs.",
          "url": "https://contralabs.com"
        }
      ],
      "areas": [
        "creative",
        "vision",
        "coding",
        "computer-use"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "This is expert input, not crowd labor.",
          "url": "https://contralabs.com/jobs"
        },
        {
          "kind": "captured",
          "quote": "Contra Labs transforms real creative workflows from our network into AI training infrastructure.",
          "url": "https://contralabs.com"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "cortex-ai",
      "name": "Cortex AI",
      "website": "https://cortexrobot.ai/",
      "what": "Provides real-world egocentric and robot data and human-in-the-loop evals to robotics AI labs.",
      "says": {
        "quote": "We power frontier labs developing robotics foundation models and general-purpose robots by providing the data they need:",
        "url": "https://www.ycombinator.com/companies/cortex-ai"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Building the world's most diverse real-world, real-workplace, and industry-scale egocentric + robot dataset, accelerating our progress toward physical intelligence.",
          "url": "https://cortexrobot.ai/"
        },
        {
          "kind": "human-data",
          "quote": "real-world deployments with remote operators who recover robots when they fail, capturing data that feeds back into training and continuously improves models.",
          "url": "https://www.ycombinator.com/companies/cortex-ai"
        },
        {
          "kind": "evals",
          "quote": "the best evals happen in the real world, with our operators providing human-in-the-loop support to capture recovery data",
          "url": "https://www.ycombinator.com/companies/cortex-ai"
        },
        {
          "kind": "annotation",
          "quote": "every frame with hand and body pose, depth, and subtask annotations.",
          "url": "https://cortexrobot.ai/"
        }
      ],
      "areas": [
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "real-workplace human video with hand/body pose, depth, and subtask labels.",
          "url": "https://www.ycombinator.com/companies/cortex-ai"
        },
        {
          "kind": "experts",
          "quote": "We're hiring full-time Robot Operators to power our real-world data operations.",
          "url": "https://cortexrobot.ai/careers/robot-operator"
        }
      ],
      "buysData": {
        "quote": "workplaces get paid to host data-collection and evaluation sessions, while labs access the in-the-wild data that truly matters.",
        "url": "https://www.ycombinator.com/companies/cortex-ai"
      },
      "hq": null,
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Lucas Ngoo, Cortex AI has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/cortex-ai"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "cua",
      "name": "Cua",
      "website": "https://cua.ai",
      "what": "Sells shared desktops for teams and agents, plus computer-use environments and tasks for training and evaluating agents.",
      "says": {
        "quote": "Train and evaluate agents on real computer-use tasks.",
        "url": "https://cua.ai/data"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Cua has 4K+ environments and 400K+ tasks spanning desktop apps, cross-app workflows, browser surfaces, and generated state variations.",
          "url": "https://cua.ai/data"
        },
        {
          "kind": "evals",
          "quote": "Train and evaluate agents on real computer-use tasks.",
          "url": "https://cua.ai/data"
        },
        {
          "kind": "datasets",
          "quote": "Selected scopes can include trajectories, annotations, and human-reviewed golden runs.",
          "url": "https://cua.ai/data"
        },
        {
          "kind": "annotation",
          "quote": "Selected scopes can include trajectories, annotations, and human-reviewed golden runs.",
          "url": "https://cua.ai/data"
        },
        {
          "kind": "human-data",
          "quote": "Selected scopes can include trajectories, annotations, and human-reviewed golden runs.",
          "url": "https://cua.ai/data"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "science",
        "creative",
        "games"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "Cua has 4K+ environments and 400K+ tasks spanning desktop apps, cross-app workflows, browser surfaces, and generated state variations.",
          "url": "https://cua.ai/data"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "cuebench",
      "name": "CueBench",
      "website": "https://cuebench.dev",
      "what": "Builds reinforcement learning environments and benchmarks for training AI models on scientific reasoning and performance engineering.",
      "says": {
        "quote": "RL environments for scientific reasoning and performance engineering.",
        "url": "https://cuebench.dev"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "CueBench builds and operates reinforcement-learning environments and benchmarks.",
          "url": "https://cuebench.dev/security"
        },
        {
          "kind": "evals",
          "quote": "CueBench builds and operates reinforcement-learning environments and benchmarks.",
          "url": "https://cuebench.dev/security"
        }
      ],
      "areas": [
        "science"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2026 by Neel Gadde, Rishan Hemrajani, and Dillon Mehta, CueBench has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/cuebench"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Neel Gadde, Rishan Hemrajani, and Dillon Mehta, CueBench has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/cuebench"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "dare-labs",
      "name": "DARE Labs",
      "website": "https://www.darelabs.net/",
      "what": "Sells Cerebro, a data refinery that turns operational data into training datasets, RL environments and evals for AI labs and operators.",
      "says": {
        "quote": "The data refinery that turns raw operational data into training-grade datasets: SFT data, RL data, and RL environments graded on real outcomes.",
        "url": "https://www.darelabs.net/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Cerebro is now the data refinery that turns raw operational data into training-grade datasets: SFT data, RL data, and RL environments graded on real outcomes.",
          "url": "https://www.darelabs.net/"
        },
        {
          "kind": "rl-environments",
          "quote": "RL environments graded on real outcomes, and hard benchmarks of how well frontier models perform your actual work, rescored on every model release.",
          "url": "https://www.darelabs.net/products"
        },
        {
          "kind": "evals",
          "quote": "Auto-curated SFT and RL datasets, generated RL gyms, and domain evals graded on real outcomes instead of proxies.",
          "url": "https://www.darelabs.net/about"
        },
        {
          "kind": "annotation",
          "quote": "DaaS automatically filters, normalizes, and labels data for fraud vs. legitimate activity while anonymizing sensitive financial information.",
          "url": "https://www.darelabs.net/about-cerebro"
        },
        {
          "kind": "simulation",
          "quote": "Integrates 3D depth data, tactile sensors, human demonstrations. DaaS generates 3D object masks & grasp points. GaaS creates varied warehouse simulations.",
          "url": "https://www.darelabs.net/about-cerebro"
        }
      ],
      "areas": [
        "finance",
        "medical",
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Turn the operational data organizations already hold into models, evals, and decision tools that do their actual work.",
          "url": "https://www.darelabs.net/"
        },
        {
          "kind": "synthetic",
          "quote": "GaaS builds synthetic transaction environments to test fraud detection models against evolving scam tactics.",
          "url": "https://www.darelabs.net/about-cerebro"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "datacurve",
      "name": "Datacurve",
      "website": "https://datacurve.ai",
      "what": "Provides coding data, RL environments, benchmarks and other training data to foundation model labs and large enterprises.",
      "says": {
        "quote": "Datacurve provides the frontier coding data that powers the world's most advanced models.",
        "url": "https://datacurve.ai/careers/growth-operations"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Durable environments for measuring agentic capabilities where real work happens: long horizons, natural instructions, realistic tools, domain-specific judgment, and partial progress.",
          "url": "https://datacurve.ai/products"
        },
        {
          "kind": "datasets",
          "quote": "Prebuilt datasets, curated for signal, reviewed for quality, and structured to drop into your training stack without translation work.",
          "url": "https://datacurve.ai/products"
        },
        {
          "kind": "human-data",
          "quote": "Demonstrations that set the right behavioral prior for models, captured through bespoke tooling that lets experts work naturally while preserving the operating judgment behind the work.",
          "url": "https://datacurve.ai/products"
        },
        {
          "kind": "evals",
          "quote": "We build benchmarks that capture task-faithful, domain-sensitive lift: progress that shows up in how the model performs, not just how the number moves.",
          "url": "https://datacurve.ai/products"
        },
        {
          "kind": "annotation",
          "quote": "We solve the data problem with our gamified annotation platform that attracts the best engineers to come and solve fun coding problems.",
          "url": "https://www.ycombinator.com/companies/datacurve"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Datacurve provides expert quality code data at scale from highly skilled software engineers.",
          "url": "https://www.ycombinator.com/companies/datacurve"
        },
        {
          "kind": "captured",
          "quote": "Full traces of expert execution, including tool calls, checks, pivots, and recoveries, so agents learn the shape of real work, not just the finished answer.",
          "url": "https://datacurve.ai/products"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2024 by Serena Ge and Charley Lee, Datacurve has 25 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/datacurve"
      },
      "founded": {
        "value": 2024,
        "quote": "Founded in 2024 by Serena Ge and Charley Lee, Datacurve has 25 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/datacurve"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "dataloop",
      "name": "Dataloop",
      "website": "https://dataloop.ai",
      "what": "Offers a data management and annotation platform for preparing visual data for machine and deep learning.",
      "says": {
        "quote": "Data annotation on DDOE refers to the process of labeling datasets (images, video, audio, text, geo-spacial data, lidar, etc.) to train and validate machine learning models.",
        "url": "https://docs.dataloop.ai/docs/data-annotation-overview"
      },
      "sells": [
        {
          "kind": "annotation",
          "quote": "Data annotation on DDOE refers to the process of labeling datasets (images, video, audio, text, geo-spacial data, lidar, etc.) to train and validate machine learning models.",
          "url": "https://docs.dataloop.ai/docs/data-annotation-overview"
        }
      ],
      "areas": [
        "vision"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "dataocean-ai",
      "name": "DataOcean AI",
      "website": "https://dataoceanai.com",
      "what": "Sells off-the-shelf AI datasets and data collection and labeling services to AI enterprises and academic institutes.",
      "says": {
        "quote": "We empower more than 900 AI enterprises with our high-quality off-the-shelf datasets, and magnificent data collection & labeling services.",
        "url": "https://dataoceanai.com/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We empower more than 900 AI enterprises with our high-quality off-the-shelf datasets, and magnificent data collection & labeling services.",
          "url": "https://dataoceanai.com/"
        },
        {
          "kind": "annotation",
          "quote": "We empower more than 900 AI enterprises with our high-quality off-the-shelf datasets, and magnificent data collection & labeling services.",
          "url": "https://dataoceanai.com/"
        }
      ],
      "areas": [
        "speech",
        "vision",
        "multilingual",
        "finance",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Only IR videos and images were captured.",
          "url": "https://dataoceanai.com/"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Spokane, United States",
        "quote": "100 N Howard St STE-R Spokane, WA 99201",
        "url": "https://dataoceanai.com/contact-us"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "datoric",
      "name": "Datoric",
      "website": "https://www.datoric.com/",
      "what": "Provides custom training data for voice models, robotics and world models.",
      "says": {
        "quote": "Datoric provides custom, security-first training data for voice models, robotics, and world models.",
        "url": "https://www.ycombinator.com/companies/datoric"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Datoric develops custom datasets for voice models, robotics, and world models, treating research, collection, verification, and production as one continuous process.",
          "url": "https://www.ycombinator.com/companies/datoric"
        },
        {
          "kind": "human-data",
          "quote": "Build-to-spec voice-preference data with human rankings and spoken rationales, scoped to the target model and evaluation criteria.",
          "url": "https://www.datoric.com/datasets"
        },
        {
          "kind": "annotation",
          "quote": "Egocentric video of household and workplace tasks, with task, action, and object labels.",
          "url": "https://www.datoric.com/"
        }
      ],
      "areas": [
        "robotics",
        "computer-use",
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We collect this data through more than 300,000 active contributors who record conversations, everyday tasks, and other project-specific activities through our private apps.",
          "url": "https://www.ycombinator.com/companies/datoric"
        },
        {
          "kind": "captured",
          "quote": "Contributors capture egocentric task video, human manipulation, or robot teleoperation to the specification.",
          "url": "https://www.datoric.com/how-it-works"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Datoric has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/datoric"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Jeffrey Lin and Nikhil Reddy",
        "url": "https://www.ycombinator.com/companies/datoric"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "david-ai",
      "name": "David AI",
      "website": "https://withdavid.ai",
      "what": "Builds audio datasets and evaluations for speech and conversational AI labs and companies.",
      "says": {
        "quote": "We are an audio data research company. ... Our datasets are used by Fortune 100 companies and research labs that work with speech recognition, translation, synthesis, and conversational AI.",
        "url": "https://withdavid.ai"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Our flagship English dataset consists of channel-separated, natural two-speaker conversations spanning a wide range of topics.",
          "url": "https://withdavid.ai"
        },
        {
          "kind": "human-data",
          "quote": "A collection of expert conversations across a range of domains.",
          "url": "https://withdavid.ai"
        },
        {
          "kind": "evals",
          "quote": "The DAI-S2S-ST Human Preference Leaderboard measures which voice models people prefer in open-ended conversations, not just whether a task was completed.",
          "url": "https://research.withdavid.ai/blog/st-s2s-eval"
        }
      ],
      "areas": [
        "speech",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "A collection of expert conversations across a range of domains.",
          "url": "https://withdavid.ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2024,
        "quote": "David AI was founded in 2024 by a team of former Scale AI engineers and operators.",
        "url": "https://www.ycombinator.com/companies/david-ai/jobs"
      },
      "readOn": "2026-10-16"
    },
    {
      "id": "deccan-ai",
      "name": "Deccan AI",
      "website": "https://deccan.ai",
      "what": "Sells post-training data, RL environments and evaluations to frontier AI labs, and agent software to enterprises.",
      "says": {
        "quote": "Post-training data and RL envs for frontier labs.",
        "url": "https://deccan.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Train and evaluate your agents in enterprise-grade RL environments",
          "url": "https://www.deccan.ai/rl-gym"
        },
        {
          "kind": "datasets",
          "quote": "Audio, video, visual, and doc intelligence datasets for mid/post training and evals.",
          "url": "https://deccan.ai"
        },
        {
          "kind": "human-data",
          "quote": "Deccan SMEs evaluate every step of  agent execution and correct reasoning, tools  and tools parameters for SFT data pairs.",
          "url": "https://www.deccan.ai/rl-gym"
        },
        {
          "kind": "evals",
          "quote": "Independent benchmarks that test the limits of frontier models",
          "url": "https://deccan.ai"
        },
        {
          "kind": "annotation",
          "quote": "Labelled image and video datasets for computer vision and multimodal model training.",
          "url": "https://deccan.ai"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "multilingual",
        "speech",
        "vision",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Every scenario written by a domain SME from workflows they run daily.",
          "url": "https://deccan.ai"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Mountain View, United States",
        "quote": "257 Castro Street, Suite #107, Mountain View, California 94041, US",
        "url": "https://deccan.ai"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "deeptune",
      "name": "Deeptune",
      "website": "https://deeptune.com",
      "what": "Builds training gyms, simulated software environments where AI agents practice work, for frontier research labs.",
      "says": {
        "quote": "We build training gyms for frontier research labs.",
        "url": "https://deeptune.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We've built 100s of gyms that mimic popular software like Slack and",
          "url": "https://deeptune.com"
        },
        {
          "kind": "datasets",
          "quote": "All gyms come with problems, datasets, and infrastructure.",
          "url": "https://deeptune.com"
        }
      ],
      "areas": [
        "coding"
      ],
      "madeFrom": [
        {
          "kind": "replica",
          "quote": "We've built 100s of gyms that mimic popular software like Slack and",
          "url": "https://deeptune.com"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "defined-ai",
      "name": "Defined.ai",
      "website": "https://defined.ai",
      "what": "Sells AI training datasets, plus data collection, annotation and evaluation services, to AI builders and enterprises.",
      "says": {
        "quote": "We empower AI builders with high-quality, ethically sourced training data backed by expert services that help you collect, annotate, and evaluate data to customize, scale, and succeed.",
        "url": "https://defined.ai"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Access off-the-shelf ethically sourced and high-quality datasets across audio, image, video, text, and multimodal formats",
          "url": "https://defined.ai"
        },
        {
          "kind": "human-data",
          "quote": "Tap into a crowd of 1.6M+ global experts spanning 150+ countries and 500+ languages, dialects, and locales",
          "url": "https://defined.ai"
        },
        {
          "kind": "annotation",
          "quote": "From custom data collection and annotation to evaluation, we support your entire AI lifecycle, helping you move from development to deployment seamlessly.",
          "url": "https://defined.ai"
        },
        {
          "kind": "evals",
          "quote": "End-to-end AI training data, from collection to evaluation",
          "url": "https://defined.ai"
        }
      ],
      "areas": [
        "enterprise",
        "finance",
        "medical",
        "robotics",
        "speech",
        "vision",
        "games",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Tap into a crowd of 1.6M+ global experts spanning 150+ countries and 500+ languages, dialects, and locales",
          "url": "https://defined.ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "dissei",
      "name": "Dissei",
      "website": "https://dissei.ai",
      "what": "Builds financial training data, evaluations and RL environments for AI labs and enterprise teams.",
      "says": {
        "quote": "Carefully built training data, evaluations and environments, grounded in real financial work, for AI labs and enterprise teams.",
        "url": "https://dissei.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "A dated financial world, its tools and graders, ready to run inside your infrastructure.",
          "url": "https://dissei.ai/products.html"
        },
        {
          "kind": "datasets",
          "quote": "Worked examples and training records built from real financial decisions, reviewed by practitioners.",
          "url": "https://dissei.ai/products.html"
        },
        {
          "kind": "human-data",
          "quote": "People who have carried the risk lead the financial standard and author the work.",
          "url": "https://data.dissei.credit/faq"
        },
        {
          "kind": "evals",
          "quote": "Assess financial judgment against fixed criteria, with evidence behind each score.",
          "url": "https://dissei.ai/products.html"
        }
      ],
      "areas": [
        "finance"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "Secondary licensed origination can add further cases under its own provenance and rights.",
          "url": "https://data.dissei.credit/faq"
        },
        {
          "kind": "experts",
          "quote": "People who have carried the risk lead the financial standard and author the work.",
          "url": "https://data.dissei.credit/faq"
        }
      ],
      "buysData": {
        "quote": "Dissei lending is the primary source, with secondary licensed origination added under its own rights.",
        "url": "https://data.dissei.credit/faq"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "dmodel",
      "name": "dmodel",
      "website": "https://dmodel.ai",
      "what": "Builds RL environments for open-ended interpretability tasks and partners with frontier labs on interpretability and alignment research.",
      "says": {
        "quote": "d_model is a fundamental AI research lab partnering with frontier labs to turn their models into capable interpretability and alignment researchers.",
        "url": "https://dmodel.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We build RL environments for open-ended interpretability tasks.",
          "url": "https://dmodel.ai"
        }
      ],
      "areas": [
        "coding"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "can easily scale up and train on thousands of examples generated from scratch",
          "url": "https://dmodel.ai/nullability-gentle/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2024,
        "quote": "Founded in 2024 by Daniel Moon and Anish Tondwalkar, d_model has 20 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/d_model"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "e2f",
      "name": "e2f",
      "website": "https://e2f.com",
      "what": "Sells AI hallucination detection and an RL evaluation platform to teams assessing AI-generated answers, plus AI training and evaluation services.",
      "says": {
        "quote": "Deterministic trust layer and reinforcement learning engine, purpose-built to assess accuracy of AI generated answers, continuously at scale.",
        "url": "https://e2f.com"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Pre- and Post- training and evaluation, benchmarking and RLHF.",
          "url": "https://e2f.com"
        },
        {
          "kind": "human-data",
          "quote": "Pre- and Post- training and evaluation, benchmarking and RLHF.",
          "url": "https://e2f.com"
        }
      ],
      "areas": [
        "legal",
        "medical",
        "multilingual"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "edotenv",
      "name": "EdotEnv",
      "website": "https://edotenv.com",
      "what": "Builds RL environments and evaluation benchmarks from financial market data for frontier AI labs and academic groups.",
      "says": {
        "quote": "We work with frontier AI labs and academic groups building research harnesses, evaluation benchmarks, and post-training environments.",
        "url": "https://edotenv.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "To get there, we build RL environments at scale from financial market data that increase in difficulty, so agents can continuously hillclimb.",
          "url": "https://edotenv.com/"
        },
        {
          "kind": "evals",
          "quote": "To this end, Alpha Autoresearch evaluates the ability of agents to generate predictive signals (alpha) from noisy market data.",
          "url": "https://edotenv.com/blog/alpha-autoresearch"
        }
      ],
      "areas": [
        "finance"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2026 by Rui Wang and Michael Zhang, EdotEnv has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/edotenv"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Rui Wang and Michael Zhang, EdotEnv has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/edotenv"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "eigent-ai",
      "name": "Eigent AI",
      "website": "https://eigent.ai",
      "what": "Makes Eigent, an open-source desktop agent, and Toolathlon-GYM, a tool-use environment for training and evaluating LLM agents.",
      "says": {
        "quote": "Eigent Open Source Cowork is a desktop multi-agent workforce that connects to your context and can control the browser and desktop apps to automate real work.",
        "url": "https://www.eigent.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We introduce Toolathlon-GYM, a large-scale, self-contained environment with 503 tasks, 25 MCP servers, and a rich mock database.",
          "url": "https://www.eigent.ai/blog/toolathlon-gym-large-scale-long-horizon-environments-for-tool-use-agents"
        },
        {
          "kind": "datasets",
          "quote": "This dataset applies the same format at larger scale, producing a substantially bigger and more diverse task pool for training and evaluation.",
          "url": "https://www.eigent.ai/blog/toolathlon-gym-large-scale-long-horizon-environments-for-tool-use-agents"
        },
        {
          "kind": "evals",
          "quote": "Training and evaluating LLM agents on real-world tool use is hard.",
          "url": "https://www.eigent.ai/blog/toolathlon-gym-large-scale-long-horizon-environments-for-tool-use-agents"
        }
      ],
      "areas": [
        "computer-use",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "public",
          "quote": "Kaggle OULAD (Open University Learning Analytics Dataset) for Learning Management System data, Kaggle HR Analytics for enterprise HR data",
          "url": "https://www.eigent.ai/blog/toolathlon-gym-large-scale-long-horizon-environments-for-tool-use-agents"
        },
        {
          "kind": "synthetic",
          "quote": "Data is derived from or simulated after real-world sources",
          "url": "https://www.eigent.ai/blog/toolathlon-gym-large-scale-long-horizon-environments-for-tool-use-agents"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "emulated",
      "name": "Emulated",
      "website": "https://emulated.so",
      "what": "Builds post-training datasets, reinforcement-learning environments and benchmarks for software engineering and AI R&D, for AI model developers.",
      "says": {
        "quote": "Emulated is a team of researchers and engineers building post-training datasets for software engineering and AI R&D.",
        "url": "https://emulated.so/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We provide ready-to-use datasets across software engineering, machine learning engineering, autonomous research, and model post-training.",
          "url": "https://emulated.so/"
        },
        {
          "kind": "evals",
          "quote": "The benchmark evaluates agents across 80 tasks organized into the five categories above.",
          "url": "https://emulated.so/sdab"
        }
      ],
      "areas": [
        "coding"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "enact",
      "name": "Enact",
      "website": "https://enact.company",
      "what": "Generates targeted robotics training data and in-house evaluations for teams building and deploying robot models.",
      "says": {
        "quote": "Enact generates the data that makes robotics models useful in the real world.",
        "url": "http://enact.company/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Enact creates the physical environments to benchmark and hill climb robotics models on high-value tasks.",
          "url": "https://www.ycombinator.com/companies/enact"
        },
        {
          "kind": "datasets",
          "quote": "Enact turns a policy’s failures into targeted recovery datasets for teams training robotics models.",
          "url": "https://www.ycombinator.com/launches/SaG-enact-post-training-infrastructure-for-robotics-models"
        },
        {
          "kind": "human-data",
          "quote": "We roll out policies on tasks, identify the states where they fail, and generate targeted demonstration and recovery datasets.",
          "url": "http://enact.company/"
        },
        {
          "kind": "evals",
          "quote": "We evaluate model performance in-house to verify that the data addresses those failures.",
          "url": "http://enact.company/"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Roll out an existing model on a physical robot.",
          "url": "https://www.ycombinator.com/launches/SaG-enact-post-training-infrastructure-for-robotics-models"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Enact has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/enact"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Govind Chada and James Stevens, Enact has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/enact"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "encord",
      "name": "Encord",
      "website": "https://encord.com",
      "what": "Sells data annotation, expert data collection, and evaluation services, plus a data platform, to AI teams building physical and multimodal AI.",
      "says": {
        "quote": "Access expert annotators, domain specialists, and managed collection pipelines - fully integrated with the Encord platform.",
        "url": "https://encord.com"
      },
      "sells": [
        {
          "kind": "annotation",
          "quote": "Annotate and review images, videos, audio, text and DICOM files faster using custom workflows, AI-assisted automation, and human-in-the-loop evaluation in the Encord platform.",
          "url": "http://encord.com/annotate"
        },
        {
          "kind": "human-data",
          "quote": "Vetted domain experts matched to your task, across physical AI, multimodal systems, and LLMs.",
          "url": "https://encord.com"
        },
        {
          "kind": "datasets",
          "quote": "Training-ready data for physical AI, collected by in-field operators and at our teleoperation facilities.",
          "url": "https://encord.com"
        },
        {
          "kind": "evals",
          "quote": "Orchestrate RLHF, rubric-based evaluation, and pairwise comparison for models running in production.",
          "url": "https://encord.com"
        }
      ],
      "areas": [
        "vision",
        "robotics",
        "medical",
        "enterprise",
        "finance",
        "speech"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Vetted domain experts matched to your task, across physical AI, multimodal systems, and LLMs.",
          "url": "https://encord.com"
        },
        {
          "kind": "captured",
          "quote": "Three pillars: data collection (in-field, in-lab, teleoperation, and synthetic capture), annotation with built-in QA, and evaluation services covering model output review, red-teaming, preference ranking, and benchmarking.",
          "url": "https://encord.com/data-services"
        },
        {
          "kind": "synthetic",
          "quote": "Three pillars: data collection (in-field, in-lab, teleoperation, and synthetic capture), annotation with built-in QA, and evaluation services covering model output review, red-teaming, preference ranking, and benchmarking.",
          "url": "https://encord.com/data-services"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "exponential",
      "name": "exponential",
      "website": "https://expsec.ai/",
      "what": "Builds multi-agent security environments for RL training and adaptive benchmarking of AI agents.",
      "says": {
        "quote": "We build multi-agent security environments for RL training and adaptive benchmarking.",
        "url": "https://expsec.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We build multi-agent security environments for RL training and adaptive benchmarking.",
          "url": "https://expsec.ai/"
        },
        {
          "kind": "datasets",
          "quote": "Security environments, adaptive red-teaming evaluations, and adversarial training data for AI agents.",
          "url": "https://expsec.ai/"
        },
        {
          "kind": "evals",
          "quote": "We use the resulting agents to benchmark how other AI agents operate under realistic adversarial conditions.",
          "url": "https://expsec.ai/"
        }
      ],
      "areas": [
        "security",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "The initial release contains expert-authored evaluation items for Europe and India, with goal-specific rubrics and no model responses or attack trajectories.",
          "url": "https://huggingface.co/ExpSec"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "fleet",
      "name": "Fleet",
      "website": "https://fleetai.com",
      "what": "Creates simulated worlds and real-world challenges to shape AI behavior, and works with frontier labs on post-training.",
      "says": {
        "quote": "Fleet1 creates simulated worlds and real-world challenges to better understand and shape the behavior of artificial intelligences.",
        "url": "https://www.fleetai.com/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "building benchmarks that expose where frontier models break",
          "url": "https://www.fleetai.com/careers/mts-synthetic-data"
        },
        {
          "kind": "datasets",
          "quote": "Build reusable and reliable systems for producing high-quality synthetic data, tasks, and verifiers",
          "url": "https://www.fleetai.com/careers/mts-synthetic-data"
        },
        {
          "kind": "rl-environments",
          "quote": "Build reusable and reliable systems for producing high-quality synthetic data, tasks, and verifiers",
          "url": "https://www.fleetai.com/careers/mts-synthetic-data"
        },
        {
          "kind": "human-data",
          "quote": "Collaborate closely with environment and research teams, frontier labs, and Fleet’s expert network to translate research needs and model failures into shared system capabilities.",
          "url": "https://www.fleetai.com/careers/mts-synthetic-data"
        }
      ],
      "areas": [
        "coding",
        "security",
        "legal",
        "finance",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Collaborate closely with environment and research teams, frontier labs, and Fleet’s expert network to translate research needs and model failures into shared system capabilities.",
          "url": "https://www.fleetai.com/careers/mts-synthetic-data"
        },
        {
          "kind": "synthetic",
          "quote": "Build reusable and reliable systems for producing high-quality synthetic data, tasks, and verifiers",
          "url": "https://www.fleetai.com/careers/mts-synthetic-data"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "foundry-ai",
      "name": "Foundry AI",
      "website": "https://foundryhq.ai",
      "what": "Builds RL-ready datasets, private evaluations, and regression environments for frontier AI models and agents.",
      "says": {
        "quote": "FoundryAI captures how experts decide, correct, and verify, then turns that judgment into custom and off-the-shelf RL-ready data, private evaluations, and regression environments.",
        "url": "https://foundryhq.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Off-the-shelf task families with environments, expert rubrics, corrected trajectories, preference pairs, RL-ready verifiers, and held-out evaluation splits.",
          "url": "https://foundryhq.ai/"
        },
        {
          "kind": "datasets",
          "quote": "FoundryAI produces custom data programs and off-the-shelf RL-ready dataset suites, with expert-verified trajectories, preference data, private evaluations, and regression environments matched to the behavior you need to improve.",
          "url": "https://foundryhq.ai/data-development"
        },
        {
          "kind": "human-data",
          "quote": "Domain-matched SFT trajectories, preference pairs, expert corrections, multimodal examples, and execution evidence built around the behavior you need to improve.",
          "url": "https://foundryhq.ai/"
        },
        {
          "kind": "evals",
          "quote": "Uncontaminated task suites, expert rubrics, hidden verifiers, and failure analysis grounded in the workflows your model is expected to perform.",
          "url": "https://foundryhq.ai/"
        }
      ],
      "areas": [
        "coding",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Run expert sessions, capture trajectories, verify outputs, and adjudicate edge cases.",
          "url": "https://foundryhq.ai/data-development"
        },
        {
          "kind": "captured",
          "quote": "Models and experts work in controlled environments while complete trajectories are recorded.",
          "url": "https://foundryhq.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "general-reasoning",
      "name": "General Reasoning",
      "website": "https://gr.inc",
      "what": "Hosts reinforcement learning environments for training and evaluating agents, serving frontier labs and AI-first startups.",
      "says": {
        "quote": "OpenReward runs thousands of Dockerized reinforcement learning environments, serving frontier labs and AI-first startups that train and evaluate agents.",
        "url": "https://www.gr.inc/careers/mts-infrastructure"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Environments are hosted as simple API endpoints that you can use for both training and evaluation, with no infrastructure to set up.",
          "url": "https://www.gr.inc/openreward"
        },
        {
          "kind": "evals",
          "quote": "Run evaluations on agentic environments without worrying about the infrastructure.",
          "url": "https://www.gr.inc/openreward"
        }
      ],
      "areas": [
        "coding",
        "math",
        "science",
        "finance",
        "games"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": {
        "value": "London, United Kingdom",
        "quote": "A London research lab building the infrastructure for AI that reasons, plans, and acts together over long horizons.",
        "url": "https://www.gr.inc/careers"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "good-start-labs",
      "name": "Good Start Labs",
      "website": "https://www.goodstartlabs.com",
      "what": "Builds game-based learning environments, benchmarks and human gameplay data for AI labs training and evaluating models.",
      "says": {
        "quote": "We build interactive learning environments from curated games with unique, verifiable rewards, so models get measurably better at hard, useful work.",
        "url": "https://www.goodstartlabs.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Interactive worlds with clean, verifiable rewards, where models learn by doing.",
          "url": "https://goodstartlabs.com/products"
        },
        {
          "kind": "datasets",
          "quote": "Games are training data for more capable models",
          "url": "https://goodstartlabs.com/products"
        },
        {
          "kind": "human-data",
          "quote": "Text and image data from real people playing, captured at the source with clear provenance.",
          "url": "https://goodstartlabs.com/products"
        },
        {
          "kind": "evals",
          "quote": "Public, transparent measurement of model capability, so every gain is real and reproducible.",
          "url": "https://goodstartlabs.com/products"
        },
        {
          "kind": "annotation",
          "quote": "Full engine access unlocks unparalleled labeling accuracy.",
          "url": "https://www.goodstartlabs.com"
        }
      ],
      "areas": [],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Recorded human gameplay with synchronized signals, so models learn how the world looks and how actions move through it.",
          "url": "https://goodstartlabs.com/benchmarks"
        },
        {
          "kind": "synthetic",
          "quote": "Synthetic and real human data at scale, including continuous human feedback from live games.",
          "url": "https://goodstartlabs.com/products"
        },
        {
          "kind": "customer",
          "quote": "We dig into your gameplay data, find what is genuinely useful to AI labs, and help you license and monetize it.",
          "url": "https://goodstartlabs.com/for-game-studios"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "gradient-health",
      "name": "Gradient Health",
      "website": "https://gradienthealth.io",
      "what": "Provides de-identified multimodal medical data through its Atlas platform to AI developers and research teams.",
      "says": {
        "quote": "Connecting healthcare data partners with AI developers through secure access to multimodal medical data.",
        "url": "https://gradienthealth.io"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Gradient Health is a medical technology company providing foundation model-scale medical imaging datasets to accelerate equitable health innovations.",
          "url": "https://gradienthealth.io/blog/the-rajpurkar-lab-and-gradient-health-unveil-rexgradient-160k-a-large-scale-multi-institutional-chest-x-ray-dataset-to-accelerate-medical-ai-research"
        }
      ],
      "areas": [
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "We constantly look for new data partners with hard-to-access or in-demand data to add to the millions of patient journeys we already have.",
          "url": "https://gradienthealth.io"
        }
      ],
      "buysData": {
        "quote": "Data partners receive a revenue share, every time their data is used by our AI developer customers.",
        "url": "https://gradienthealth.io/data-partners"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "gray-swan-ai",
      "name": "Gray Swan AI",
      "website": "https://www.grayswan.ai/",
      "what": "Provides automated adversarial testing, red teaming and runtime AI security to enterprises and frontier AI labs.",
      "says": {
        "quote": "We provide automated adversarial testing, continuous red teaming, and runtime AI security for enterprises deploying AI and the frontier labs building it.",
        "url": "https://www.grayswan.ai/about"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Gray Swan’s research team develops the adversarial benchmarks, evaluation frameworks, and safety methodologies that frontier labs use to measure model risk before release.",
          "url": "https://www.grayswan.ai/solutions/for-model-builders/adversarial-evaluation"
        },
        {
          "kind": "datasets",
          "quote": "High-risk and regulated domains: evaluation and datasets where standard benchmarks don't apply",
          "url": "https://www.grayswan.ai/solutions/for-model-builders/adversarial-evaluation"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "15,000+ red teamers generating continuous adversarial data",
          "url": "https://www.grayswan.ai/solutions/for-model-builders/adversarial-evaluation"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "halluminate",
      "name": "Halluminate",
      "website": "https://halluminate.ai",
      "what": "Builds RL environments and benchmarks that train and evaluate frontier AI models on financial services and consulting knowledge work.",
      "says": {
        "quote": "Halluminate is a data research lab building benchmarks and RL environments to push the frontier of AI in knowledge work.",
        "url": "https://halluminate.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Built and scaled RL environments with four of the five leading closed-source U.S. AI labs",
          "url": "https://www.ycombinator.com/companies/halluminate"
        },
        {
          "kind": "evals",
          "quote": "We present Westworld Finance Diligence Bench, a benchmark of 88 problems that evaluates AI agents on a complete company acquisition due-diligence process.",
          "url": "https://halluminate.ai/"
        }
      ],
      "areas": [
        "finance",
        "computer-use"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Every problem is written and reviewed by practicing finance deal professionals and verified by a mixed pipeline of agentic, discrete, and binary verifiers.",
          "url": "https://halluminate.ai/"
        }
      ],
      "buysData": {
        "quote": "Turn your company’s data and workflows into a new revenue opportunity.",
        "url": "https://halluminate.ai/data-partnerships"
      },
      "hq": {
        "value": "San Francisco, USA",
        "quote": "Founded in 2024 by Wyatt Marshall and Jerry Wu, Halluminate has 5 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/halluminate"
      },
      "founded": {
        "value": 2024,
        "quote": "Founded in 2024 by Wyatt Marshall and Jerry Wu, Halluminate has 5 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/halluminate"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "handshake",
      "name": "Handshake AI",
      "website": "https://joinhandshake.com/research/ai/",
      "what": "Provides expert human model validation, RL environments and evaluations to AI labs, drawing on a network of graduate-level experts.",
      "says": {
        "quote": "Handshake AI is our new service to provide expert human model validation and technology systems for frontier AI development.",
        "url": "https://joinhandshake.com/blog/our-team/introducing-handshake-ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "MIT-founded AI team and tech deepen Handshake investment in frontier AI research in RL environments, evals, and human data.",
          "url": "https://joinhandshake.com/blog/our-team/handshake-acquires-cleanlab"
        },
        {
          "kind": "evals",
          "quote": "ATLAS tests whether AI can actually do the job.",
          "url": "https://joinhandshake.com/research/benchmarks"
        },
        {
          "kind": "human-data",
          "quote": "The experts train large language models using their academic expertise",
          "url": "https://joinhandshake.com/blog/our-team/introducing-handshake-ai"
        },
        {
          "kind": "annotation",
          "quote": "producing data on our own annotation platform",
          "url": "https://joinhandshake.com/blog/our-team/introducing-handshake-ai"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "legal",
        "science",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Our experts are sourced directly from universities",
          "url": "https://joinhandshake.com/blog/our-team/introducing-handshake-ai"
        },
        {
          "kind": "licensed",
          "quote": "Handshake AI buys that data and works with top AI labs to understand how real companies do work–all with data that is scoped, de-identified, and confidential.",
          "url": "https://joinhandshake.com/ai/data-partnerships"
        }
      ],
      "buysData": {
        "quote": "Handshake AI buys that data and works with top AI labs to understand how real companies do work–all with data that is scoped, de-identified, and confidential.",
        "url": "https://joinhandshake.com/ai/data-partnerships"
      },
      "hq": null,
      "founded": {
        "value": 2014,
        "quote": "In 2014, we founded Handshake to change how, where, and why the next generation of talent builds their career",
        "url": "https://joinhandshake.com/about"
      },
      "readOn": "2026-10-16"
    },
    {
      "id": "hillclimb",
      "name": "Hillclimb",
      "website": "https://hillclimb.com",
      "what": "Designs and sells RL environments and training data, including math problem sets, to frontier AI labs.",
      "says": {
        "quote": "We work with frontier AI labs to help train their agents to become AI research scientists",
        "url": "https://www.ycombinator.com/companies/hillclimb"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Starting with math, we’ve formed the world’s densest cluster of IMO medalists, lean experts, and other math talent to design RL environments for frontier labs.",
          "url": "https://www.ycombinator.com/companies/hillclimb"
        },
        {
          "kind": "datasets",
          "quote": "Explore curated datasets of mathematical problems and AI model evaluations",
          "url": "https://data.hillclimb.ing/"
        },
        {
          "kind": "human-data",
          "quote": "we’re building a community of the world’s brightest minds to create the next generation of training data",
          "url": "https://www.ycombinator.com/companies/hillclimb"
        },
        {
          "kind": "evals",
          "quote": "Curated set of 50 original mathematical problems designed by Hillclimb researchers to evaluate reasoning capabilities.",
          "url": "https://data.hillclimb.ing/"
        }
      ],
      "areas": [
        "math"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Built the densest cluster of elite math talent in the industry (IMO medalists, Putnam Top 50, Lean formalization experts, etc.)",
          "url": "https://www.ycombinator.com/companies/hillclimb"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, USA",
        "quote": "Founded in 2025 by Jun Park, hillclimb has 4 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/hillclimb"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Jun Park, hillclimb has 4 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/hillclimb"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "hiloop",
      "name": "hiloop",
      "website": "https://hiloop.ai",
      "what": "Runs automated research campaigns that improve customers' models and agents through post-training, optimization, and synthetic RL environments.",
      "says": {
        "quote": "We turn a measurable target into thousands of directed experiments, and return a better model or algorithm.",
        "url": "https://hiloop.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "The environments an agent trains against, generated and scored by the loop instead of hand-built one at a time.",
          "url": "https://hiloop.ai"
        }
      ],
      "areas": [],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "The environments an agent trains against, generated and scored by the loop instead of hand-built one at a time.",
          "url": "https://hiloop.ai"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2026 by Thomas Boser and Karan Brar, Autoresearch, Inc. has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/hiloop"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Thomas Boser and Karan Brar, Autoresearch, Inc. has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/hiloop"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "hud",
      "name": "HUD",
      "website": "https://hud.ai",
      "what": "Provides a platform and open-source SDK for building environments and tasks, evaluating agents against them and training models on the results.",
      "says": {
        "quote": "HUD is a platform for building environments.",
        "url": "https://docs.hud.ai/v6"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Introduction to HUD: define an environment once, run any agent against it for evaluation, and train models on the trajectories your tasks produce.",
          "url": "https://docs.hud.ai/v6"
        },
        {
          "kind": "evals",
          "quote": "Introduction to HUD: define an environment once, run any agent against it for evaluation, and train models on the trajectories your tasks produce.",
          "url": "https://docs.hud.ai/v6"
        }
      ],
      "areas": [],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-16"
    },
    {
      "id": "human-archive",
      "name": "Human Archive",
      "website": "https://www.humanarchive.ai/",
      "what": "Collects human video and sensor data and sells datasets to frontier AI labs for physical AI and robotics.",
      "says": {
        "quote": "Leading frontier AI labs trust us for their video, sensor, robot deployment, image, and audio datasets.",
        "url": "https://www.ycombinator.com/companies/human-archive"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Leading frontier AI labs trust us for their video, sensor, robot deployment, image, and audio datasets.",
          "url": "https://www.ycombinator.com/companies/human-archive"
        },
        {
          "kind": "human-data",
          "quote": "Human Archive is currently working alongside leading AI labs to scale human video/sensor data and deploy robotics.",
          "url": "https://www.humanarchive.ai/"
        },
        {
          "kind": "annotation",
          "quote": "We provide annotations and metadata including environment and scene descriptions, high-level task descriptions, task-aligned atomic action labels, hand tracking, object segmentation, SLAM (extrinsic and intrinsics), and 3D pose reconstruction.",
          "url": "https://www.ycombinator.com/companies/human-archive"
        }
      ],
      "areas": [
        "robotics",
        "vision",
        "speech"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "The Service lets adult contributors record first-person video of everyday tasks and household activities, submit those recordings to us, and earn payouts for submissions that we approve.",
          "url": "https://www.humanarchive.ai/privacy.html"
        },
        {
          "kind": "experts",
          "quote": "we’ve signed national level partnerships to scale our contributor network to 50,000+ people.",
          "url": "https://www.ycombinator.com/companies/human-archive"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2026 by Rushil Agarwal, Samay Maini, Raj Patel, and Shloke Patel, Human Archive has 100 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/human-archive"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Rushil Agarwal, Samay Maini, Raj Patel, and Shloke Patel, Human Archive has 100 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/human-archive"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "humyn-labs",
      "name": "Humyn Labs",
      "website": "https://humynlabs.ai",
      "what": "Collects and annotates custom AI training datasets across voice, image, video and sensor data for AI model builders.",
      "says": {
        "quote": "Humyn Labs builds custom AI training datasets across voice, image, video, and sensor modalities collected and annotated by verified domain experts with multi-layer quality control.",
        "url": "https://humynlabs.ai/solutions/ai-training-data"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Humyn Labs builds custom AI training datasets across voice, image, video, and sensor modalities collected and annotated by verified domain experts with multi-layer quality control.",
          "url": "https://humynlabs.ai/solutions/ai-training-data"
        },
        {
          "kind": "human-data",
          "quote": "Your training data is produced by vetted domain experts with on-chain reputation, not anonymous crowd workers.",
          "url": "https://humynlabs.ai/solutions/ai-training-data"
        },
        {
          "kind": "annotation",
          "quote": "Domain experts label it with 3D boxes, segmentation, tracking, and scene classification.",
          "url": "https://humynlabs.ai/solutions/physical-ai-data"
        },
        {
          "kind": "evals",
          "quote": "We handle synchronization, validation, processing, QC, and benchmark evaluation",
          "url": "https://humynlabs.ai"
        }
      ],
      "areas": [
        "speech",
        "vision",
        "robotics",
        "multilingual",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Your training data is produced by vetted domain experts with on-chain reputation, not anonymous crowd workers.",
          "url": "https://humynlabs.ai/solutions/ai-training-data"
        },
        {
          "kind": "captured",
          "quote": "Verified field teams capture multi-sensor data in the real world.",
          "url": "https://humynlabs.ai/solutions/physical-ai-data"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "huzzle-labs",
      "name": "Huzzle Labs",
      "website": "https://labs.huzzle.com",
      "what": "Builds RL environments and evaluations for frontier AI labs, plus custom models for enterprises.",
      "says": {
        "quote": "We build RL environments for frontier AI labs, and the evaluations and custom models that put AI to work inside the enterprise.",
        "url": "https://labs.huzzle.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Realistic, long-horizon environments for code, computer-use, and enterprise workflows that challenge SOTA models.",
          "url": "https://labs.huzzle.com/rl-environments"
        },
        {
          "kind": "evals",
          "quote": "We build evaluations with frontier AI labs, and work with enterprises across Europe.",
          "url": "https://labs.huzzle.com/enterprise"
        },
        {
          "kind": "datasets",
          "quote": "a human data and evaluation services business providing reinforcement learning environments, RLHF datasets, and contextual evaluations to AI laboratories and enterprises.",
          "url": "https://www.huzzle.com/privacy-policy"
        },
        {
          "kind": "human-data",
          "quote": "the Huzzle Labs platform (annotation, evaluation, and human data services for AI laboratories and enterprises)",
          "url": "https://www.huzzle.com/terms-of-service"
        },
        {
          "kind": "annotation",
          "quote": "the Huzzle Labs platform (annotation, evaluation, and human data services for AI laboratories and enterprises)",
          "url": "https://www.huzzle.com/terms-of-service"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Long-horizon RL environments for code, computer-use and enterprise workflows that challenge SOTA models, built on a network of 300,000+ domain experts.",
          "url": "https://labs.huzzle.com/rl-environments"
        },
        {
          "kind": "customer",
          "quote": "We train on your data and deploy on your own infrastructure, so your data and the models built on it stay yours.",
          "url": "https://labs.huzzle.com/enterprise"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "London, UK",
        "quote": "85 Great Portland Street, London, UK",
        "url": "https://www.huzzle.com/privacy-policy"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "idler",
      "name": "Idler",
      "website": "https://idler.ai",
      "what": "Builds evals, RL environments and off-the-shelf datasets that AI labs use to measure and train their models.",
      "says": {
        "quote": "We build the evals and environments that the world's leading frontier labs use to measure and train their models.",
        "url": "http://ycombinator.com/companies/idler"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Hill-climb any benchmark, for any model, in any domain using our RL environments.",
          "url": "https://www.ycombinator.com/launches/SxT-idler-push-the-frontier-of-your-model"
        },
        {
          "kind": "datasets",
          "quote": "Now, we're offering off-the-shelf datasets across an even broader set of capabilities",
          "url": "http://ycombinator.com/companies/idler"
        },
        {
          "kind": "evals",
          "quote": "We make bespoke evaluations for your model.",
          "url": "https://idler.ai/collections"
        },
        {
          "kind": "simulation",
          "quote": "A stateful simulator of retail operations.",
          "url": "https://idler.ai/collections/shelflife-e-sim/documentation"
        }
      ],
      "areas": [
        "coding",
        "legal",
        "finance",
        "security",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "we work with domain experts to understand these discrepancies and construct detailed, information-dense environments.",
          "url": "https://idler.ai/collections/shelflife/documentation"
        },
        {
          "kind": "customer",
          "quote": "we present a collection of 200 environments drawn from the company's operational data over the past decade",
          "url": "https://idler.ai/collections/shelflife/documentation"
        }
      ],
      "buysData": {
        "quote": "You have 10-100 employees and you’d like to supplement your revenue by licensing your codebase or de-identified business data.",
        "url": "http://ycombinator.com/companies/idler"
      },
      "hq": {
        "value": "San Francisco",
        "quote": "idler has 12 employees based in San Francisco",
        "url": "http://ycombinator.com/companies/idler"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Ivan Chub, Nalu Concepcion, and Tony Goss",
        "url": "http://ycombinator.com/companies/idler"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "imerit",
      "name": "iMerit",
      "website": "https://imerit.ai",
      "what": "Sells expert human data, model evaluation, and annotation services to foundation model labs and enterprise AI teams.",
      "says": {
        "quote": "evaluation, training data, alignment, and safety for frontier labs and enterprise AI teams.",
        "url": "https://imerit.ai"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Custom-sourced global subject experts for multimodal supervised fine-tuning, RLHF, red teaming, alignment and validation.",
          "url": "https://imerit.ai"
        },
        {
          "kind": "evals",
          "quote": "Multidomain Tests & Benchmarks: Challenging puzzles in specialized fields and languages",
          "url": "https://imerit.ai"
        },
        {
          "kind": "annotation",
          "quote": "Annotation and validation for Computer Vision, Audio, LiDAR, and Sensor Fusion with guaranteed high quality and long-term consistency",
          "url": "https://imerit.ai"
        },
        {
          "kind": "datasets",
          "quote": "Original corpus of 1,000 Chain of Thought math problems, including step-wise corrections and annotation, intersecting 60+ disparate subdomains of formal and applied mathematics.",
          "url": "https://imerit.ai"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "medical",
        "science",
        "math",
        "robotics",
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Custom-sourced global subject experts for multimodal supervised fine-tuning, RLHF, red teaming, alignment and validation.",
          "url": "https://imerit.ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "incalmo",
      "name": "Incalmo",
      "website": "https://www.incalmo.ai",
      "what": "Builds RL environments and security evaluations for frontier labs, large security vendors, and enterprises.",
      "says": {
        "quote": "RL environments that task models with",
        "url": "https://www.incalmo.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "RL environments that task models with",
          "url": "https://www.incalmo.ai/"
        },
        {
          "kind": "datasets",
          "quote": "building systems that generate cyber data at scale",
          "url": "https://www.incalmo.ai/careers"
        }
      ],
      "areas": [
        "security"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "innodata",
      "name": "Innodata",
      "website": "https://innodata.com/",
      "what": "Sells AI training datasets, data annotation, model evaluations and RL environments to companies building AI models.",
      "says": {
        "quote": "Seven of the world’s largest tech companies trust Innodata for fine-tuning and red teaming.",
        "url": "https://innodata.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Custom reinforcement learning environments for training AI agents to reason, recover from errors, and complete complex, long-horizon tasks.",
          "url": "https://innodata.com/data-annotation/"
        },
        {
          "kind": "datasets",
          "quote": "Pre-built AI training datasets for LLMs. Scalable, expert-curated data for model training, evaluation, and reasoning performance.",
          "url": "https://innodata.com/data-annotation/"
        },
        {
          "kind": "human-data",
          "quote": "Rely on human experts to improve hallucinations and edge-cases with ongoing feedback.",
          "url": "https://innodata.com/data-annotation/"
        },
        {
          "kind": "evals",
          "quote": "Evaluate model capabilities and uncover failure modes across cultural alignment, complex interactions, and safety.",
          "url": "https://innodata.com/data-annotation/"
        },
        {
          "kind": "annotation",
          "quote": "Deliver accurate, domain-specific data annotation across all data types",
          "url": "https://innodata.com/data-annotation/"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "medical",
        "robotics",
        "vision",
        "speech",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Image, video, text, document, speech, and sensor data labeled by trained subject-matter experts across 120+ languages",
          "url": "https://innodata.com/data-annotation/"
        },
        {
          "kind": "captured",
          "quote": "Real-world training data for robotics, VLA models, world models, wearables, and embodied AI.",
          "url": "https://innodata.com/data-annotation/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "invisible",
      "name": "Invisible Technologies",
      "website": "https://invisibletech.ai",
      "what": "Sells AI training datasets, RL environments, evaluations and data annotation to companies that build AI models.",
      "says": {
        "quote": "Model-ready, off-the-shelf data Datasets spanning professional work, code, and customer interactions. Available now.",
        "url": "https://invisibletech.ai/data-lab/data"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Designed for your priorities Train agents around your goals through realistic tasks that reflect real-world tools, data, uncertainty, and decision-making.",
          "url": "https://invisibletech.ai/solutions/rl-environments"
        },
        {
          "kind": "datasets",
          "quote": "Model-ready, off-the-shelf data Datasets spanning professional work, code, and customer interactions. Available now.",
          "url": "https://invisibletech.ai/data-lab/data"
        },
        {
          "kind": "human-data",
          "quote": "Our experts define task logic, design reward rubrics, and annotate trajectories.",
          "url": "https://invisibletech.ai/solutions/rl-environments"
        },
        {
          "kind": "evals",
          "quote": "Run rigorous evaluation for quality, safety, and accuracy.",
          "url": "https://invisibletech.ai"
        },
        {
          "kind": "annotation",
          "quote": "Go from idea to labeled data in minutes–no specs, no handoffs, just a custom annotation interface powered by natural language and instantly deployable to your team or our expert workforce. Built for any modality, any language.",
          "url": "https://invisibletech.ai/scale-training-data"
        }
      ],
      "areas": [
        "coding",
        "finance",
        "legal",
        "enterprise",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We source elite domain experts for on-demand projects.",
          "url": "https://invisibletech.ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-16"
    },
    {
      "id": "kled",
      "name": "Kled",
      "website": "https://kled.ai",
      "what": "Licenses datasets of video, music, transcriptions and other human-made data to AI labs, governments and research institutions.",
      "says": {
        "quote": "Kled provides premium, unbiased data from top rightsholders to train AI models.",
        "url": "https://www.kled.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We curate high-quality datasets, including video, music, and transcriptions, enriched with metadata to ensure accurate, ethical AI training and optimal model performance.",
          "url": "https://www.kled.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Powering leading AI labs, governments, and research institutions with verified human data.",
          "url": "https://www.kled.ai/"
        },
        {
          "kind": "evals",
          "quote": "Eval sets are continuously built to surface model failures and identify the specific data required to close those gaps.",
          "url": "https://www.kled.ai/blog/kled%E2%80%99s-opt-in-human-data-network-at-scale"
        },
        {
          "kind": "annotation",
          "quote": "Once sensitive data is removed and originality is confirmed, fine-tuned multimodal models are run to label the data.",
          "url": "https://www.kled.ai/blog/kled%E2%80%99s-opt-in-human-data-network-at-scale"
        }
      ],
      "areas": [
        "vision",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "Kled provides premium, unbiased data from top rightsholders to train AI models.",
          "url": "https://www.kled.ai/"
        },
        {
          "kind": "experts",
          "quote": "Powering the world’s leading AI companies, governments, and research institutions with data sourced from verified contributors.",
          "url": "https://www.kled.ai/"
        },
        {
          "kind": "captured",
          "quote": "All collected through distributed human-in-the-loop capture.",
          "url": "https://www.kled.ai/blog/kled%E2%80%99s-opt-in-human-data-network-at-scale"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "labelbox",
      "name": "Labelbox",
      "website": "https://labelbox.com",
      "what": "Builds the environments and expert data frontier AI labs train on, and AI agents for engineering and research teams.",
      "says": {
        "quote": "Labelbox builds the environments frontier labs train AI on, and the agents engineering and research teams hand their backlog to.",
        "url": "https://labelbox.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Labelbox builds the environments frontier labs train AI on, and the agents engineering and research teams hand their backlog to.",
          "url": "https://labelbox.com"
        },
        {
          "kind": "human-data",
          "quote": "Frontier AI labs train on Labelbox environments and expert data.",
          "url": "https://labelbox.com"
        },
        {
          "kind": "evals",
          "quote": "After seeing the impact of using Labelbox internally, Google Cloud selected it to deliver LLM evaluation as a managed solution inside Vertex AI.",
          "url": "https://labelbox.com/customers/"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "robotics",
        "speech",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Labelbox's platform produces expert-graded preference signal across customizable dimensions like instruction following, verbosity, and relevance.",
          "url": "https://labelbox.com/customers/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "latch",
      "name": "Latch",
      "website": "https://latch.bio",
      "what": "Builds biosecurity benchmarks, audits and red-teaming that evaluate AI models for biological risk, for frontier AI labs.",
      "says": {
        "quote": "Helping frontier AI labs evaluate and audit their systems for biosecurity risk.",
        "url": "https://latch.bio/security"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Benchmarks and evaluations of AI models across the biosecurity-relevant capabilities that matter, from advancing biological threats to strengthening biodefense.",
          "url": "https://latch.bio/security"
        },
        {
          "kind": "datasets",
          "quote": "A 30M cell observational atlas spanning 150 diseases, 200 tissues and 27 technologies",
          "url": "https://blog.latch.bio/p/latchbio-now-provides-lots-of-data"
        },
        {
          "kind": "annotation",
          "quote": "helps them perform tasks like count matrix construction, cell typing and metadata harmonization with greater efficiency and accuracy",
          "url": "https://blog.latch.bio/p/latchbio-now-provides-lots-of-data"
        }
      ],
      "areas": [
        "science",
        "medical",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "public",
          "quote": "The data used for these evaluations is from publicly available manuscripts, case studies, or databases.",
          "url": "https://latch.bio"
        },
        {
          "kind": "experts",
          "quote": "Our first 30M cell atlas represents thousands of hours of human labor - the combined efforts of the Pythia Biosciences, Miraomics and LatchBio curation teams.",
          "url": "https://blog.latch.bio/p/latchbio-now-provides-lots-of-data"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "lightwheel",
      "name": "Lightwheel",
      "website": "https://lightwheel.ai",
      "what": "Provides SimReady simulation assets, egocentric human data, and evaluation platforms to robotics and Physical AI teams.",
      "says": {
        "quote": "Lightwheel is an Physical AI infrastructure company, building SimReady assets, egocentric human data, and evaluation platforms that power robots to learn and deploy at scale.",
        "url": "https://lightwheel.ai"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "The first batch of EgoSuite-Open100K is live on the Hugging Face Hub, open for academic research and commercial training, with the remainder rolling out in stages.",
          "url": "https://lightwheel.ai/media/egosuite-open100k"
        },
        {
          "kind": "human-data",
          "quote": "Lightwheel builds SimReady assets, synthetic and egocentric human data, and evaluation platforms for Embodied AI.",
          "url": "https://lightwheel.ai"
        },
        {
          "kind": "evals",
          "quote": "Industrial-grade simulation evaluation platform challenging frontier robotics models and world models.",
          "url": "https://www.lightwheel.ai/robofinals"
        },
        {
          "kind": "annotation",
          "quote": "It includes hand pose and body pose annotations, wrist-camera coverage for part of the dataset, and a complimentary event-level semantic annotation add-on for selected subsets.",
          "url": "https://lightwheel.ai/media/egosuite-open100k"
        },
        {
          "kind": "simulation",
          "quote": "Powered by OpenUSD, Lightwheel has developed standardized pipelines to produce large volumes of physics-consistent simulation assets and environments while preserving alignment with real-world physics.",
          "url": "https://lightwheel.ai/media/simready"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Lightwheel has built a Physical Measurement Factory designed to systematically measure how real objects behave.",
          "url": "https://lightwheel.ai/media/simready"
        },
        {
          "kind": "experts",
          "quote": "It was built through a globally distributed workforce of tens of thousands and a continuous, standardized collection process that keeps capture conventions consistent across the dataset.",
          "url": "https://lightwheel.ai/media/egosuite-open100k"
        },
        {
          "kind": "synthetic",
          "quote": "Lightwheel builds SimReady assets, synthetic and egocentric human data, and evaluation platforms for Embodied AI.",
          "url": "https://lightwheel.ai"
        },
        {
          "kind": "public",
          "quote": "The first batch of EgoSuite-Open100K is live on the Hugging Face Hub, open for academic research and commercial training, with the remainder rolling out in stages.",
          "url": "https://lightwheel.ai/media/egosuite-open100k"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "lightwrk",
      "name": "LightWrk",
      "website": "https://lightwrk.ai",
      "what": "Sells reasoning-rich annotations, RL evals, and preference data from vetted experts to teams training robot and world models.",
      "says": {
        "quote": "Reasoning-rich annotations, RL evals, and preference data from vetted experts.",
        "url": "https://lightwrk.ai/about"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Expert human reasoning data for physical AI",
          "url": "https://lightwrk.ai/about"
        },
        {
          "kind": "annotation",
          "quote": "Reasoning-rich annotations, RL evals, and preference data from vetted experts.",
          "url": "https://lightwrk.ai/about"
        },
        {
          "kind": "evals",
          "quote": "Reasoning-rich annotations, RL evals, and preference data from vetted experts.",
          "url": "https://lightwrk.ai/about"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Reasoning-rich annotations, RL evals, and preference data from vetted experts.",
          "url": "https://lightwrk.ai/about"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "luel",
      "name": "Luel",
      "website": "https://luel.ai",
      "what": "Sells rights-cleared multimodal training datasets, collected by a global contributor network, for building next-generation AI.",
      "says": {
        "quote": "Luel is the marketplace, and the global contributor network behind it, that ships the rights-cleared multimodal datasets the next generation of AI is being built on.",
        "url": "https://www.luel.ai/blog/launching-luel-yc-w26"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Explore, license, and analyze production-ready datasets or brief us on a custom build we scope from scratch.",
          "url": "https://www.luel.ai/marketplace"
        },
        {
          "kind": "human-data",
          "quote": "One of the largest seed rounds in Y Combinator's history, powering Luel's worldwide contributor network that supplies human data to the world's leading AI labs.",
          "url": "https://www.luel.ai/resources/blog/luel-seed-31m-general-catalyst-lightspeed"
        },
        {
          "kind": "evals",
          "quote": "Luel against named systems on a 50-sentence Egyptian Arabic set, scored by a panel of eight independent recognizers.",
          "url": "https://www.luel.ai/lab"
        },
        {
          "kind": "annotation",
          "quote": "exclusive, custom datasets built to spec with custom annotations",
          "url": "https://www.luel.ai/resources/blog/luel-seed-31m-general-catalyst-lightspeed"
        }
      ],
      "areas": [
        "speech",
        "computer-use",
        "robotics",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "a global, vetted contributor network that produces datasets to a buyer's spec",
          "url": "https://www.luel.ai/blog/launching-luel-yc-w26"
        },
        {
          "kind": "captured",
          "quote": "Luel operates a worldwide contributor network, hundreds of thousands of people across almost 100 countries, who are paid to record, upload, and license slices of their everyday lives.",
          "url": "https://www.luel.ai/resources/blog/luel-seed-31m-general-catalyst-lightspeed"
        },
        {
          "kind": "licensed",
          "quote": "third-party listings, where external labs post their own datasets on Luel's marketplace for a revenue share",
          "url": "https://www.luel.ai/resources/blog/luel-seed-31m-general-catalyst-lightspeed"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Bay Area, United States",
        "quote": "The company was founded in late 2025 by William Namgyal and Inigo Lenderking, is part of Y Combinator's W26 cohort, and is based in the Bay Area.",
        "url": "https://www.luel.ai/resources/blog/luel-seed-31m-general-catalyst-lightspeed"
      },
      "founded": {
        "value": 2025,
        "quote": "Luel was founded in 2025 by William Namgyal, CEO (18), and Inigo Lenderking, COO (19).",
        "url": "https://www.luel.ai/resources/blog/luel-seed-31m-general-catalyst-lightspeed"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "lxt",
      "name": "LXT",
      "website": "https://lxt.ai",
      "what": "Provides AI training data collection, annotation, evaluation and fine-tuning services to AI companies.",
      "says": {
        "quote": "Customized, scalable data pipelines for AI training data, model fine-tuning, and evaluation – across text, speech, image, and video, supporting GenAI, NLP, speech, vision, and multimodal systems.",
        "url": "https://www.lxt.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Access pre-built, curated datasets ready for immediate use.",
          "url": "https://www.lxt.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Our experts craft instruction–response pairs and apply RLHF to align model outputs with human preferences.",
          "url": "https://www.lxt.ai/"
        },
        {
          "kind": "evals",
          "quote": "Evaluation pipelines built for safety, fairness, and compliance",
          "url": "https://www.lxt.ai/"
        },
        {
          "kind": "annotation",
          "quote": "Transform raw data into machine-readable insights with expert human labeling and classification.",
          "url": "https://www.lxt.ai/"
        }
      ],
      "areas": [
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Our managed global crowd deliver accurate datasets for even the most complex AI use cases.",
          "url": "https://www.lxt.ai/"
        },
        {
          "kind": "captured",
          "quote": "Multimodal data capture: audio, image, video, text",
          "url": "https://www.lxt.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "macrodata-labs",
      "name": "Macrodata Labs",
      "website": "https://macrodata.co/",
      "what": "Sells automated annotation, QA and 3D hand-motion extraction for robot and egocentric video to robotics teams.",
      "says": {
        "quote": "Turn robot and egocentric recordings into automated QA, timestamped subtasks, and 3D hand trajectories using our proprietary models and pipelines.",
        "url": "https://macrodata.co/"
      },
      "sells": [
        {
          "kind": "annotation",
          "quote": "Our pipeline finds where each subtask starts and ends, producing timestamped subtask annotations for training long-horizon robot policies.",
          "url": "https://macrodata.co/"
        },
        {
          "kind": "evals",
          "quote": "Send a small video sample for a free evaluation.",
          "url": "https://macrodata.co/"
        }
      ],
      "areas": [
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Send us a small video sample. We’ll annotate it for free and return the results for your team to evaluate.",
          "url": "https://macrodata.co/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "magma",
      "name": "Magma",
      "website": "https://magmahq.ai",
      "what": "Packages the interaction data of companies' AI agents and licenses it to AI labs and researchers for a revenue share.",
      "says": {
        "quote": "Magma packages it and licenses it.",
        "url": "https://magmahq.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Magma anonymizes the data, turns it into complex training environments, and handles licensing to AI labs.",
          "url": "https://www.ycombinator.com/companies/magma"
        },
        {
          "kind": "datasets",
          "quote": "We package production interaction data: traces, corrections, real workflows.",
          "url": "https://magmahq.ai"
        }
      ],
      "areas": [],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "We package your agent's interaction data for AI labs.",
          "url": "https://magmahq.ai"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2026 by Ramtin Shahzanian, Magma has 1 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/magma"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Ramtin Shahzanian, Magma has 1 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/magma"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "maingen",
      "name": "Maingen",
      "website": "https://www.maingen.ai/",
      "what": "Builds RL environments and simulations of industrial operations, such as running solar portfolios, so frontier AI labs can train models.",
      "says": {
        "quote": "Maingen builds RL environments for industrials (a $5 trillion slice of the US economy) so frontier labs can train models that actually run factories.",
        "url": "https://www.ycombinator.com/companies/maingen"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Maingen builds RL environments for industrials",
          "url": "https://www.ycombinator.com/companies/maingen"
        },
        {
          "kind": "evals",
          "quote": "SolarBench assesses whether frontier AI agents can resolve long-horizon, multi-domain tasks as the on-call engineer for a portfolio of solar farms.",
          "url": "https://solarbench.maingen.ai/"
        },
        {
          "kind": "simulation",
          "quote": "Maingen builds simulations for industrial operations",
          "url": "https://www.ycombinator.com/companies/maingen"
        }
      ],
      "areas": [],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "SolarBench scenarios are drawn from hundreds of hours with the people who run solar for a living.",
          "url": "https://solarbench.maingen.ai/"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2026 by Phillip Yan and David Yang, Maingen has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/maingen"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Phillip Yan and David Yang, Maingen has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/maingen"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "markov",
      "name": "Markov",
      "website": "https://markovstudios.com",
      "what": "Builds expert computer-use datasets and environments for training computer-use AI, sold to frontier AI labs.",
      "says": {
        "quote": "Markov builds expert computer-use datasets for frontier AI labs.",
        "url": "https://www.ycombinator.com/companies/markov"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Markov builds expert computer-use datasets for frontier AI labs.",
          "url": "https://www.ycombinator.com/companies/markov"
        },
        {
          "kind": "rl-environments",
          "quote": "Data + environments for training computer-use AI.",
          "url": "https://markovstudios.com"
        },
        {
          "kind": "evals",
          "quote": "AutoCAD Bench tests whether multimodal agents can translate a dimensioned reference drawing into a correct, editable DWG. It is a fixed 50-task benchmark spanning 2D drafting and 3D reconstruction.",
          "url": "https://www.markovstudios.com/research/autocad-bench"
        },
        {
          "kind": "human-data",
          "quote": "We work with a vetted network of expert contributors to collect high-quality computer-use data across",
          "url": "https://www.ycombinator.com/companies/markov"
        },
        {
          "kind": "annotation",
          "quote": "Markov builds the data infrastructure for computer-use AI, collecting synchronized screen recordings, mouse/keyboard actions, and annotations from real workflows.",
          "url": "https://www.ycombinator.com/companies/markov"
        }
      ],
      "areas": [
        "computer-use",
        "enterprise",
        "finance",
        "games"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We work with a vetted network of expert contributors to collect high-quality computer-use data across",
          "url": "https://www.ycombinator.com/companies/markov"
        },
        {
          "kind": "captured",
          "quote": "and turn it into data models can learn from.",
          "url": "https://markovstudios.com"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Bengaluru, India",
        "quote": "Founded in 2026 by Harish Ashok and Dev Mandal, Markov has 2 employees based in Bengaluru.",
        "url": "https://www.ycombinator.com/companies/markov"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Harish Ashok and Dev Mandal, Markov has 2 employees based in Bengaluru.",
        "url": "https://www.ycombinator.com/companies/markov"
      },
      "readOn": "2026-10-16"
    },
    {
      "id": "maxinsights",
      "name": "Maxinsights",
      "website": "https://www.maxinsights.ai/",
      "what": "Provides egocentric video data collection, annotation, motion capture, simulation and evaluation for robotics and physical AI model builders.",
      "says": {
        "quote": "Providing large-scale multimodal data collection, hand tracking, upper-body & whole-body motion capture, tactile force signals, simulation, and data evaluation for generalist robotics, Vision-Language-Action (VLA) models, and world action models (WAM).",
        "url": "https://www.maxinsights.ai/blog/Beyond_1M_Hours"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Our data goes beyond clean demonstrations, exposing models to real-world ambiguity, recovery scenarios, complex coordination, and long-horizon tasks.",
          "url": "https://www.maxinsights.ai/"
        },
        {
          "kind": "evals",
          "quote": "Providing large-scale multimodal data collection, hand tracking, upper-body & whole-body motion capture, tactile force signals, simulation, and data evaluation for generalist robotics, Vision-Language-Action (VLA) models, and world action models (WAM).",
          "url": "https://www.maxinsights.ai/blog/Beyond_1M_Hours"
        },
        {
          "kind": "annotation",
          "quote": "The premier physical AI & annotated egocentric data vendor in the San Francisco Bay Area.",
          "url": "https://www.maxinsights.ai/blog/Beyond_1M_Hours"
        },
        {
          "kind": "simulation",
          "quote": "Providing large-scale multimodal data collection, hand tracking, upper-body & whole-body motion capture, tactile force signals, simulation, and data evaluation for generalist robotics, Vision-Language-Action (VLA) models, and world action models (WAM).",
          "url": "https://www.maxinsights.ai/blog/Beyond_1M_Hours"
        }
      ],
      "areas": [
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "people perform tasks, sensors capture them, hardware stays calibrated, and the recording is then synchronized, validated, annotated, and converted into something a training system can consume.",
          "url": "https://www.maxinsights.ai/blog/Beyond_1M_Hours"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco Bay Area",
        "quote": "The premier physical AI & annotated egocentric data vendor in the San Francisco Bay Area.",
        "url": "https://www.maxinsights.ai/blog/Beyond_1M_Hours"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "mechanize",
      "name": "Mechanize",
      "website": "https://mechanize.work",
      "what": "Builds RL environments and evals for frontier coding agents, sold to leading AI labs.",
      "says": {
        "quote": "Mechanize is a software company that builds RL environments and sells them to the leading AI labs.",
        "url": "https://mechanize.work"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Mechanize is a software company that builds RL environments and sells them to the leading AI labs.",
          "url": "https://mechanize.work"
        },
        {
          "kind": "evals",
          "quote": "We build environments and evals for frontier coding agents.",
          "url": "https://mechanize.work"
        },
        {
          "kind": "datasets",
          "quote": "Today we’re announcing Mechanize, a startup focused on developing virtual work environments, benchmarks, and training data that will enable the full automation of the economy.",
          "url": "https://www.mechanize.work/announcing-mechanize-inc"
        },
        {
          "kind": "simulation",
          "quote": "We will achieve this by creating simulated environments and evaluations that capture the full scope of what people do at their jobs.",
          "url": "https://www.mechanize.work/announcing-mechanize-inc"
        }
      ],
      "areas": [
        "coding"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Currently, our model for producing these tasks involves each task being owned by a single engineer, who is responsible for all phases of the task lifecycle",
          "url": "https://www.mechanize.work/what-working-here-is-like"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "mecka",
      "name": "Mecka",
      "website": "https://mecka.ai",
      "what": "Builds data, evaluation and deployment infrastructure for robots and embodied AI, for robotics labs, frontier model teams and enterprises.",
      "says": {
        "quote": "Mecka builds the infrastructure for robots to learn about the real world and deploy reliably into commercial settings.",
        "url": "https://www.mecka.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Egoverse is Mecka's large-scale human interaction dataset, built from real-world first-person activity across tasks, spaces, and environments",
          "url": "https://www.mecka.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Explore human data from around the world, built for robot learning",
          "url": "https://www.mecka.ai/"
        },
        {
          "kind": "evals",
          "quote": "Start with dataset access, commission task-specific collection, run model evaluations, or build a full data program across environments.",
          "url": "https://www.mecka.ai/"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "mercor",
      "name": "Mercor",
      "website": "https://www.mercor.com/research/",
      "what": "Builds benchmarks, RL environments and large-scale human datasets for AI labs and enterprise model developers, sourced through its expert marketplace.",
      "says": {
        "quote": "We develop benchmarks, evaluation environments, and large-scale human datasets to fuel AI breakthroughs at the frontier, all through our marketplace of top-tier experts.",
        "url": "https://www.mercor.com/research/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We build reinforcement learning (RL) environments in three steps",
          "url": "https://www.mercor.com/research/"
        },
        {
          "kind": "datasets",
          "quote": "Diverse datasets spanning professional, coding, research, agentic, and multimodal tasks.",
          "url": "https://www.mercor.com/apex/off-the-shelf-data/"
        },
        {
          "kind": "human-data",
          "quote": "Mercor's talent platform mobilizes deep subject-matter experts across professional and consumer domains to produce specialized data at scale.",
          "url": "https://www.mercor.com/research/"
        },
        {
          "kind": "evals",
          "quote": "Benchmarks, peer-reviewed research, evaluation leaderboards, datasets, and open tooling for frontier AI systems.",
          "url": "https://www.mercor.com/research/"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "legal",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Every task is written by PhD researchers, practicing lawyers, and senior engineers, not crowdsourced.",
          "url": "https://www.mercor.com/apex/off-the-shelf-data/"
        },
        {
          "kind": "replica",
          "quote": "Expert-built tasks run inside high-fidelity enterprise app clones",
          "url": "https://www.mercor.com/apex/off-the-shelf-data/"
        },
        {
          "kind": "synthetic",
          "quote": "Engineers scripted synthetic logs (500 to 1,000 lines of normal operation mixed with 10-20 lines of bug symptoms)",
          "url": "https://www.mercor.com/blog/introducing-apex-swe/"
        },
        {
          "kind": "licensed",
          "quote": "We license it, package it, and pay you for it.",
          "url": "https://www.mercor.com/data"
        }
      ],
      "buysData": {
        "quote": "We license it, package it, and pay you for it.",
        "url": "https://www.mercor.com/data"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "metaphi",
      "name": "Metaphi",
      "website": "https://metaphi.ai",
      "what": "Licenses RL environments, agent trajectories and SFT datasets, and offers benchmarks for enterprise work, to AI research and frontier model developers.",
      "says": {
        "quote": "We offer RL environments, agent trajectories and SFT datasets on non-exclusive license to advance AI research, and frontier model development.",
        "url": "https://metaphi.ai/environments"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "RL Environments for post training Flash class models, constructed on private production codebases.",
          "url": "https://metaphi.ai/benchmarks/enterpriseswe"
        },
        {
          "kind": "datasets",
          "quote": "We offer RL environments, agent trajectories and SFT datasets on non-exclusive license to advance AI research, and frontier model development.",
          "url": "https://metaphi.ai/environments"
        },
        {
          "kind": "human-data",
          "quote": "the programs through which experts contribute to our datasets",
          "url": "https://metaphi.ai/privacy"
        },
        {
          "kind": "evals",
          "quote": "We evaluate frontier agents on real-world environments and private datasets.",
          "url": "https://metaphi.ai/benchmarks/enterpriseswe"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "the programs through which experts contribute to our datasets",
          "url": "https://metaphi.ai/privacy"
        },
        {
          "kind": "licensed",
          "quote": "When we license code or data from a third party to build a benchmark, that agreement governs what we receive.",
          "url": "https://metaphi.ai/privacy"
        }
      ],
      "buysData": {
        "quote": "When we license code or data from a third party to build a benchmark, that agreement governs what we receive.",
        "url": "https://metaphi.ai/privacy"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "micro1",
      "name": "micro1",
      "website": "https://micro1.ai",
      "what": "Provides expert human data, RL environments, and evaluations to AI labs and frontier robotics labs.",
      "says": {
        "quote": "We're building the infrastructure for advancing intelligence through expert human data, real-world training environments, and contextual evaluations",
        "url": "https://micro1.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Frontier evaluations and RL environments for AI Labs",
          "url": "https://micro1.ai"
        },
        {
          "kind": "evals",
          "quote": "Expert data and evaluation environments built with leading domain specialists across 100+ fields",
          "url": "https://www.micro1.ai/realm"
        },
        {
          "kind": "human-data",
          "quote": "High quality expert-demonstrated data for frontier robotics labs",
          "url": "https://micro1.ai"
        },
        {
          "kind": "datasets",
          "quote": "The environment where domain experts create, review, and deliver complex datasets across industries like healthcare, legal, finance, and more",
          "url": "https://www.micro1.ai/realm"
        }
      ],
      "areas": [
        "coding",
        "finance",
        "legal",
        "medical",
        "speech"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Expert data and evaluation environments built with leading domain specialists across 100+ fields",
          "url": "https://www.micro1.ai/realm"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "midcentury",
      "name": "Midcentury",
      "website": "https://midcentury.xyz",
      "what": "Sells egocentric human action datasets and a simulation platform for training and evaluating robots, to frontier AI labs and enterprises.",
      "says": {
        "quote": "Real-world egocentric data and high-fidelity simulation at scale for training, evaluating, and deploying robots.",
        "url": "https://www.midcentury.xyz/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "One of the largest unscripted egocentric datasets across industrial and everyday environments.",
          "url": "https://www.midcentury.xyz/"
        },
        {
          "kind": "human-data",
          "quote": "Human action data at internet scale for pretraining.",
          "url": "https://www.midcentury.xyz/"
        },
        {
          "kind": "simulation",
          "quote": "Agentic simulation platform to design, test, and train physical AI across massively parallel cloud environments.",
          "url": "https://www.midcentury.xyz/"
        },
        {
          "kind": "evals",
          "quote": "Introducing MC-PhysBench: a physics benchmark for world models.",
          "url": "https://www.midcentury.xyz/"
        },
        {
          "kind": "annotation",
          "quote": "Large-scale robotics and simulation data, paired with rich action supervision and dense annotations for the next generation of physical AI.",
          "url": "https://www.midcentury.xyz/"
        }
      ],
      "areas": [
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "2 million hours of first-person video of real manual labor, with IMU, 3D hand pose and frame-level action labels.",
          "url": "https://www.midcentury.xyz/datasets/egocentric"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "New York, USA",
        "quote": "32 Mercer St · New York, NY",
        "url": "https://www.midcentury.xyz/ego"
      },
      "founded": null,
      "readOn": "2026-10-16"
    },
    {
      "id": "mind-lab",
      "name": "Mind Lab",
      "website": "https://macaron.im/mindlab",
      "what": "Post-trains and benchmarks AI models for pharmaceutical, financial, public-sector and gaming companies, with services from data development through deployment.",
      "says": {
        "quote": "We provide pharmaceutical, financial, public-sector, and gaming companies with end-to-end services spanning Benchmarking, data development, model post-training, inference, and deployment",
        "url": "https://macaron.im/mindlab"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Rebuild your enterprise task environment under control, and set scoring logic and edge cases.",
          "url": "https://macaron.im/mindlab"
        },
        {
          "kind": "evals",
          "quote": "From dedicated eval and data through post-training, inference, and iteration, we take the full loop.",
          "url": "https://macaron.im/mindlab"
        },
        {
          "kind": "human-data",
          "quote": "Shipped work covers 27 QC indicators and more than 8,000 physician-experience traces, turning expert judgment into a skill the model can learn.",
          "url": "https://macaron.im/mindlab"
        }
      ],
      "areas": [
        "medical",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Our experts run benchmark, data, and post-training",
          "url": "https://macaron.im/mindlab"
        },
        {
          "kind": "customer",
          "quote": "Tell us the industry, the target task, and the data you already have.",
          "url": "https://macaron.im/mindlab"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "miso-motion",
      "name": "Miso Motion",
      "website": "https://misomotion.com/en/labs",
      "what": "Collects custom human demonstration data for physical AI, using its own network of service professionals.",
      "says": {
        "quote": "Collect human demonstration data to your spec inside a live service network of 130,000+ professionals.",
        "url": "https://misomotion.com/en/labs"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Collect human demonstration data to your spec inside a live service network of 130,000+ professionals.",
          "url": "https://misomotion.com/en/labs"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We run the sites, operators and rigs",
          "url": "https://misomotion.com/en/labs"
        },
        {
          "kind": "captured",
          "quote": "Operators record while doing the paid job, so capture rides a schedule that already exists.",
          "url": "https://misomotion.com/en/labs"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "monte",
      "name": "Monte",
      "website": "https://trymonte.ai/",
      "what": "Provides post-training and continual learning infrastructure, including evaluations, for enterprise teams building AI agents.",
      "says": {
        "quote": "Monte is an applied research lab building continual learning and post-training infrastructure for AI agents.",
        "url": "https://trymonte.ai/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Build evaluations around the workflows, constraints, edge cases, and standards your agent has to meet in production.",
          "url": "https://trymonte.ai/"
        },
        {
          "kind": "rl-environments",
          "quote": "and a model is evaluated against it, trained with supervised fine-tuning or reinforcement learning, then evaluated again.",
          "url": "https://docs.trymonte.ai/"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Traces, outcomes, policies, and expert judgment become the training signal that teaches your agent your work.",
          "url": "https://trymonte.ai/"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2025 by Henry Bae, Alex Fleury, and Jamin Liu, Monte has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/monte"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Henry Bae, Alex Fleury, and Jamin Liu, Monte has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/monte"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "moonlake-ai",
      "name": "Moonlake AI",
      "website": "https://moonlakeai.com",
      "what": "Builds world models and simulation and evaluation infrastructure for robotics teams to test and deploy robot policies.",
      "says": {
        "quote": "The simulation and eval layer for robot deployment.",
        "url": "http://moonlakeai.com/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "The simulation and eval layer for robot deployment.",
          "url": "http://moonlakeai.com/"
        },
        {
          "kind": "simulation",
          "quote": "Moonlake generates simulation environments with policy-level guarantees, meaning you have confidence that evaluation in sim has meaning for real-world performance.",
          "url": "https://moonlakeai.com/blog/simulation-environment-generation"
        }
      ],
      "areas": [
        "robotics",
        "computer-use",
        "games",
        "creative"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Moonlake reconstructed a data center’s workspace from walkthrough videos, prepared cable repatching in simulation, and adjusted the trajectory for a physical OpenArm test deployment.",
          "url": "http://moonlakeai.com/"
        },
        {
          "kind": "synthetic",
          "quote": "Targeted domain randomization (i.e., simulation data)",
          "url": "http://moonlakeai.com/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "noumad",
      "name": "Noumad",
      "website": "https://noumad.ai/",
      "what": "Builds reinforcement learning environments and verifiable tasks that train frontier AI models across ML, science, robotics, and cybersecurity.",
      "says": {
        "quote": "Noumad builds reinforcement learning environments that turn unsolved research problems into the verifiable tasks that train frontier AI models, across long-horizon ML, computational science, robotics, and cybersecurity.",
        "url": "https://noumad.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Noumad builds reinforcement learning environments that turn unsolved research problems into the verifiable tasks that train frontier AI models, across long-horizon ML, computational science, robotics, and cybersecurity.",
          "url": "https://noumad.ai/"
        },
        {
          "kind": "evals",
          "quote": "SecureAssemblyGym (SAG) evaluates whether an AI system can author an executable defender policy for that coupled problem.",
          "url": "https://noumad.ai/research/secure-assembly-gym"
        },
        {
          "kind": "simulation",
          "quote": "SecureAssemblyGym uses a simulated seven-degree-of-freedom fixed-base arm, a parallel gripper, a wrist depth camera, a workcell, and an assembly with ordered prerequisites.",
          "url": "https://noumad.ai/research/secure-assembly-gym"
        }
      ],
      "areas": [
        "robotics",
        "security",
        "science"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "olam-labs",
      "name": "Olam Labs",
      "website": "https://olamlabs.ai/",
      "what": "Runs multi-agent simulations and behavioral evaluations of AI models for frontier AI labs and researchers.",
      "says": {
        "quote": "We work with frontier AI labs and researchers to evaluate and train their models in complex simulated environments.",
        "url": "https://olamlabs.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We’re specialists and at the frontier of both RL and evaluations for multi-agent environments like swarms or social behavior.",
          "url": "https://olamlabs.ai/contact"
        },
        {
          "kind": "datasets",
          "quote": "Tasks and datasets to improve models in Multi-Agent Arena in both capability and safety.",
          "url": "https://olamlabs.ai/contact"
        },
        {
          "kind": "human-data",
          "quote": "datasets of top-performing humans and agents",
          "url": "https://olamlabs.ai/contact"
        },
        {
          "kind": "evals",
          "quote": "Pre-deployment evaluations for safety and capability through anonymized arena models and internal evals.",
          "url": "https://olamlabs.ai/contact"
        },
        {
          "kind": "simulation",
          "quote": "Craftsmen that research complex, high-fidelity simulations for training and evaluating frontier AI models alongside the infrastructure to test and deploy them at scale.",
          "url": "https://www.ycombinator.com/companies/olam-labs"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "games",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Model reasoning traces are evaluated against ground-truth game states using expert-curated rubrics and variance-checked grader panels",
          "url": "https://www.ycombinator.com/companies/olam-labs"
        },
        {
          "kind": "captured",
          "quote": "the infrastructure captures interaction data to score models across two primary axes in our first evaluated game, Poker",
          "url": "https://www.ycombinator.com/companies/olam-labs"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Om Buddhdev and shreshth sharma, Olam Labs has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/olam-labs"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "ooak",
      "name": "Ooak",
      "website": "https://ooakdata.com",
      "what": "Turns anonymized real company data into datasets and RL environments for frontier AI labs, enterprise AI teams and AI startups.",
      "says": {
        "quote": "We turn real company data into RL environments where AI agents learn to work in the real world.",
        "url": "http://ooakdata.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We turn real company data into RL environments where AI agents learn to work in the real world.",
          "url": "http://ooakdata.com/"
        },
        {
          "kind": "datasets",
          "quote": "We build the datasets, evaluation environments, and data pipelines that move AI agents from controlled demos to real-world production.",
          "url": "http://ooakdata.com/"
        },
        {
          "kind": "evals",
          "quote": "We build multi-step, multi-tool environments that test what matters: can your agent actually complete a workflow?",
          "url": "http://ooakdata.com/"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "Ooak Data licenses the operational and business data that defines how companies actually work, anonymizes it, and turns it into training material for frontier AI models.",
          "url": "https://ooakdata.com/partner"
        }
      ],
      "buysData": {
        "quote": "Ooak Data licenses real business data, anonymizes it, and builds AI training environments.",
        "url": "https://ooakdata.com/partner"
      },
      "hq": {
        "value": "Wilmington, United States",
        "quote": "Principal place of business: 108 W. 13th Street, Suite 100, Wilmington, Delaware 19801, United States",
        "url": "https://ooakdata.com/legal"
      },
      "founded": {
        "value": 2024,
        "quote": "Founded in 2024 by Thomas Aubry and Pierre-Louis Vouteau, Ooak Data has 5 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/ooak-data"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "opentrain-ai",
      "name": "OpenTrain AI",
      "website": "https://opentrain.ai",
      "what": "Places vetted domain experts and AI trainers with AI companies for LLM evaluation, RLHF, red teaming, and data labeling.",
      "says": {
        "quote": "Hire pre-vetted domain experts at scale for LLM evaluation, RLHF, red teaming, data labeling, and more.",
        "url": "https://opentrain.ai"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Human evaluation of model outputs, benchmark creation, A/B preference testing, and ongoing quality monitoring.",
          "url": "https://opentrain.ai"
        },
        {
          "kind": "human-data",
          "quote": "RLHF preference ranking, DPO pair generation, and reward model training data.",
          "url": "https://opentrain.ai"
        },
        {
          "kind": "datasets",
          "quote": "PhDs and professionals in mathematics, medicine, law, finance, and engineering providing expert-grade training data.",
          "url": "https://opentrain.ai"
        },
        {
          "kind": "annotation",
          "quote": "Vision, audio, video, and document annotation with cross-modal alignment across 118+ languages.",
          "url": "https://opentrain.ai"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "legal",
        "medical",
        "math",
        "finance",
        "science",
        "creative",
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We source domain experts, verify them through live proctored testing and interviews, then manage your team day-to-day inside your tools.",
          "url": "https://www.opentrain.ai/managed-service"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "paper-instruments",
      "name": "Paper Instruments",
      "website": "https://www.paperinstruments.com/",
      "what": "Builds evaluations for AI products, such as Perplexity's Advanced Deep Research, and open-source models for knowledge work.",
      "says": {
        "quote": "Paper Instruments partnered with Perplexity to develop frontier evaluations for its flagship Advanced Deep Research product.",
        "url": "https://www.paperinstruments.com/blog/tldc-perplexity-draco"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Paper Instruments partnered with Perplexity to develop frontier evaluations for its flagship Advanced Deep Research product.",
          "url": "https://www.paperinstruments.com/blog/tldc-perplexity-draco"
        },
        {
          "kind": "datasets",
          "quote": "Read Perplexity's DRACO announcement and technical report, or explore the DRACO dataset on Hugging Face.",
          "url": "https://www.paperinstruments.com/blog/tldc-perplexity-draco"
        },
        {
          "kind": "human-data",
          "quote": "We directed knowledge-work professionals to refine those rubrics around real-world priorities.",
          "url": "https://www.paperinstruments.com/blog/tldc-perplexity-draco"
        }
      ],
      "areas": [
        "finance",
        "legal",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "We built a synthetic data-generation pipeline that transformed production traffic into constrained, evaluable queries",
          "url": "https://www.paperinstruments.com/blog/tldc-perplexity-draco"
        },
        {
          "kind": "experts",
          "quote": "We directed knowledge-work professionals to refine those rubrics around real-world priorities.",
          "url": "https://www.paperinstruments.com/blog/tldc-perplexity-draco"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "pareto",
      "name": "Pareto",
      "website": "https://pareto.ai",
      "what": "Recruits domain experts to produce training data for AI model builders, and publishes benchmarks for evaluating LLMs.",
      "says": {
        "quote": "Our projects are built around the judgment, reasoning, and domain knowledge that only specialists have.",
        "url": "https://pareto.ai/experts"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "As someone training in orthopedic surgery, the intellectually stimulating projects let me creatively contribute medical knowledge for AI training.",
          "url": "https://pareto.ai/experts"
        },
        {
          "kind": "evals",
          "quote": "we built LEAPBench (Learning Efficiency in Adaptive Processes): a benchmark intentionally designed to measure the in-context learning efficiency of LLMs across iterative scientific design tasks.",
          "url": "https://pareto.ai/blog/introducing-leapbench"
        }
      ],
      "areas": [
        "science",
        "medical",
        "legal",
        "finance"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "As someone training in orthopedic surgery, the intellectually stimulating projects let me creatively contribute medical knowledge for AI training.",
          "url": "https://pareto.ai/experts"
        },
        {
          "kind": "public",
          "quote": "For each task, we trained a supervised regression model on the published literature datasets to act as a surrogate oracle.",
          "url": "https://pareto.ai/blog/introducing-leapbench"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "parsewave",
      "name": "Parsewave",
      "website": "https://parsewave.ai",
      "what": "Builds custom datasets, traces and expert evaluations for frontier AI models.",
      "says": {
        "quote": "Custom datasets, traces, expert evaluations and more for frontier models.",
        "url": "https://parsewave.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Parsewave datasets are designed for use in supervised fine-tuning, reinforcement learning from human feedback, or as benchmarks to assess model performance.",
          "url": "https://parsewave.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Every datapoint is created from scratch by senior practitioners, primarily senior software engineers alongside vetted domain specialists.",
          "url": "https://parsewave.ai/"
        },
        {
          "kind": "evals",
          "quote": "Custom datasets, traces, expert evaluations and more for frontier models.",
          "url": "https://parsewave.ai/"
        },
        {
          "kind": "annotation",
          "quote": "Produce tailored problem sets, solution traces, and reasoning annotations that match your technical requirements.",
          "url": "https://parsewave.ai/"
        }
      ],
      "areas": [
        "coding"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Every datapoint is created from scratch by senior practitioners, primarily senior software engineers alongside vetted domain specialists.",
          "url": "https://parsewave.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "patronus-ai",
      "name": "Patronus AI",
      "website": "https://patronus.ai",
      "what": "Develops training data, tasks, and simulated environments for frontier AI labs and reinforcement learning.",
      "says": {
        "quote": "Patronus AI develops high-quality training data, tasks, and simulated environments for frontier AI labs and reinforcement learning.",
        "url": "https://www.patronus.ai/rl-environments"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We build simulated applications and workflows where agents can execute complex tasks, delivered to run within your training infrastructure.",
          "url": "https://www.patronus.ai/rl-environments"
        },
        {
          "kind": "datasets",
          "quote": "Patronus AI develops high-quality training data, tasks, and simulated environments for frontier AI labs and reinforcement learning.",
          "url": "https://www.patronus.ai/rl-environments"
        },
        {
          "kind": "human-data",
          "quote": "Human-annotated traces from software and research tasks",
          "url": "https://www.patronus.ai/?trk=public_post-text"
        },
        {
          "kind": "evals",
          "quote": "Testing financial reasoning against real company documents",
          "url": "https://www.patronus.ai/?trk=public_post-text"
        },
        {
          "kind": "simulation",
          "quote": "Our Digital World Model generates interactive worlds dynamically.",
          "url": "https://www.patronus.ai/dwm"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Expert-curated questions with answers and supporting evidence",
          "url": "https://www.patronus.ai/?trk=public_post-text"
        },
        {
          "kind": "synthetic",
          "quote": "Through scalable world modeling and synthetic data generation, we aim to extend human expertise across more training experiences",
          "url": "https://www.patronus.ai/?trk=public_post-text"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "perle-ai",
      "name": "Perle AI",
      "website": "https://perle.ai",
      "what": "Supplies training data, evaluations, and expert labeling, delivered by vetted specialists, to teams building AI models.",
      "says": {
        "quote": "Bring structure to your pipeline with training data, evaluation, and labeling handled end-to-end by expert teams and a unified platform.",
        "url": "https://perle.ai/trust-center"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Access continuously refreshed, expert-labeled multimodal data designed for teams building real-world robotics systems.",
          "url": "https://perle.ai/trust-center"
        },
        {
          "kind": "human-data",
          "quote": "Reinforcement learning from human feedback staffed with experts who can defend their reward signal, with a paper trail your research lead can stand behind.",
          "url": "https://www.perle.ai/solutions/dataservices"
        },
        {
          "kind": "evals",
          "quote": "Side-by-side comparisons, capability probes, refusal calibration, bias audits, written by people who do the underlying work for a living.",
          "url": "https://www.perle.ai/solutions"
        },
        {
          "kind": "annotation",
          "quote": "Expert labeling for the cases generic vendors return blank.",
          "url": "https://www.perle.ai/solutions/dataservices"
        }
      ],
      "areas": [
        "coding",
        "robotics",
        "legal",
        "medical",
        "finance",
        "speech",
        "vision",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Our evaluators are domain experts who’ve done the work your agent is trying to imitate.",
          "url": "https://perle.ai/solutions/agentic-ai"
        },
        {
          "kind": "captured",
          "quote": "WhisperMind captures the way real hands move through real spaces, picking up a wet glass, opening a stuck drawer, finding the salt in a stranger's cupboard.",
          "url": "https://perle.ai/trust-center"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "petrarch",
      "name": "Petrarch",
      "website": "https://petrarch.co/",
      "what": "Sources, cleans and resells internal company datasets to applied AI startups and labs.",
      "says": {
        "quote": "Petrarch sources suppliers, cleans and restructures the datasets, certifies legal clearance, and resells to applied AI startups and labs.",
        "url": "https://www.ycombinator.com/companies/petrarch"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We source data including codebases, project files, and payments from bankruptcy courts, then de-identify and prepare the data for model training.",
          "url": "https://www.ycombinator.com/companies/petrarch"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "Petrarch sources suppliers, cleans and restructures the datasets, certifies legal clearance, and resells to applied AI startups and labs.",
          "url": "https://www.ycombinator.com/companies/petrarch"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, United States",
        "quote": "Petrarch has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/petrarch"
      },
      "founded": {
        "value": 2026,
        "quote": "Founded in 2026 by Samuel Hahn, Ian Lee, and Sudhish Swain, Petrarch has 3 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/petrarch"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "pixelstwo",
      "name": "Pixelstwo",
      "website": "https://pixelstwo.com/",
      "what": "Collects and sells tactile and dexterous manipulation data for robots, captured with proprietary hardware.",
      "says": {
        "quote": "High-quality, curated tactile and dexterous robotics data, captured with proprietary hardware and delivered at scale.",
        "url": "https://www.pixelstwo.com/scale"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Share the behaviors, embodiment, and volume you need. Let's discuss the right dataset.",
          "url": "https://www.pixelstwo.com/scale"
        }
      ],
      "areas": [
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Contact-rich interaction data collected with proprietary tactile hardware.",
          "url": "https://www.pixelstwo.com/scale"
        }
      ],
      "buysData": {
        "quote": "Run a business in Bengaluru or Delhi? Get paid to record your team’s everyday tasks by wearing a cap.",
        "url": "https://pixelstwo.com/apply/"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "plato",
      "name": "Plato",
      "website": "https://plato.so",
      "what": "Provides Plato VM environments, including Linux desktops, that agents operate and that testcases score.",
      "says": {
        "quote": "A computer use env is a Plato VM running a full Linux desktop (Xvfb, window manager, Chrome, real filesystem).",
        "url": "https://docs.plato.so/computer-use/quickstart"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "A testcase is a pre-defined task: prompt, target sim(s), starting artifacts, scoring config.",
          "url": "https://docs.plato.so/concepts/overview"
        }
      ],
      "areas": [
        "computer-use"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "poindexter-labs",
      "name": "Poindexter Labs",
      "website": "https://www.poindexterlabs.ai/",
      "what": "Sells expert-led data annotation and evaluations, and a collaboration platform, to frontier AI labs, governments and enterprises.",
      "says": {
        "quote": "We combine specialist expertise, research-grade standards and proprietary workflow technology to design evaluations, coordinate expert collaboration and produce reliable evidence of model capability.",
        "url": "https://www.poindexterlabs.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Poindexter Labs creates high-fidelity reasoning data and secure collaborative workflows that help frontier AI teams train, evaluate and improve advanced models.",
          "url": "https://www.poindexterlabs.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Our elite team – the Poindexters – delivers data annotation for the most complex tasks in frontier AI.",
          "url": "https://www.poindexterlabs.ai/"
        },
        {
          "kind": "evals",
          "quote": "We combine specialist expertise, research-grade standards and proprietary workflow technology to design evaluations, coordinate expert collaboration and produce reliable evidence of model capability.",
          "url": "https://www.poindexterlabs.ai/"
        },
        {
          "kind": "annotation",
          "quote": "Our elite team – the Poindexters – delivers data annotation for the most complex tasks in frontier AI.",
          "url": "https://www.poindexterlabs.ai/"
        }
      ],
      "areas": [
        "finance",
        "legal"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "They’re experts in their subject matter, with highly advanced reasoning, analysis and judgement skills.",
          "url": "https://www.poindexterlabs.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "polymath",
      "name": "Polymath",
      "website": "https://www.polymathlabs.ai",
      "what": "Builds simulation environments for agents to practice and learn through experience, for leading model labs.",
      "says": {
        "quote": "Polymath builds simulation environments for agents to practice and learn through experience.",
        "url": "https://polymathlabs.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Polymath builds simulation environments for agents to practice and learn through experience.",
          "url": "https://polymathlabs.ai/"
        },
        {
          "kind": "evals",
          "quote": "Simulation environments for training & evaluating autonomous agents.",
          "url": "https://polymathlabs.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Fellows use domain expertise to identify where current models fail, and then create high-quality training environments and data to teach AI how to improve on realistic engineering and research workflows.",
          "url": "http://fellows.polymathlabs.ai/r/software-engineer-888736"
        }
      ],
      "areas": [
        "coding",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Fellows use domain expertise to identify where current models fail, and then create high-quality training environments and data to teach AI how to improve on realistic engineering and research workflows.",
          "url": "http://fellows.polymathlabs.ai/r/software-engineer-888736"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "pre-dev",
      "name": "pre.dev",
      "website": "https://pre.dev",
      "what": "Sells verified RL coding tasks to frontier AI labs for post-training, alongside its coding agent and browser agent products.",
      "says": {
        "quote": "pre.dev Labs sells verified RL coding tasks to frontier AI labs for post-training.",
        "url": "https://pre.dev"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Hard, verified RL tasks for coding agents, built from private production code.",
          "url": "https://pre.dev/rl-environments"
        },
        {
          "kind": "evals",
          "quote": "built for your own RL or evaluation infrastructure.",
          "url": "https://pre.dev/rl-environments"
        }
      ],
      "areas": [
        "coding",
        "computer-use"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "Licensed directly from the teams that built it.",
          "url": "https://pre.dev/rl-environments"
        },
        {
          "kind": "replica",
          "quote": "Each task recreates one real commit from a licensed repo.",
          "url": "https://pre.dev/rl-environments"
        }
      ],
      "buysData": {
        "quote": "Licensed directly from the teams that built it.",
        "url": "https://pre.dev/rl-environments"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "preference-model",
      "name": "Preference Model",
      "website": "https://www.preferencemodel.com",
      "what": "Builds RL environments used to train AI models, with sample tasks in coding, games, math and image classification.",
      "says": {
        "quote": "That's why we're building RL environments for training capable, better-aligned superintelligences.",
        "url": "https://www.preferencemodel.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "That's why we're building RL environments for training capable, better-aligned superintelligences.",
          "url": "https://www.preferencemodel.com"
        }
      ],
      "areas": [
        "games",
        "math",
        "vision"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "prime-intellect",
      "name": "Prime Intellect",
      "website": "https://www.primeintellect.ai/",
      "what": "Provides hosted training, RL environments, evaluations and compute for teams that train their own AI models.",
      "says": {
        "quote": "Train, deploy, and continuously improve your own models on an integrated compute, training, inference, and sandbox stack.",
        "url": "https://www.primeintellect.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Turn any task into an RL environment. Init, develop, eval, and push with the Prime CLI.",
          "url": "https://www.primeintellect.ai/"
        },
        {
          "kind": "evals",
          "quote": "Hosted evaluations for you to benchmark the performance of your models.",
          "url": "https://www.primeintellect.ai/"
        },
        {
          "kind": "datasets",
          "quote": "SYNTHETIC-2 consists of a large set of verifiable reasoning tasks as well as reasoning traces obtained from multiple models.",
          "url": "https://www.primeintellect.ai/blog/synthetic-2"
        }
      ],
      "areas": [
        "coding",
        "science",
        "math",
        "games"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "We generate reasoning traces for all of our tasks using DeepSeek-R1-0528.",
          "url": "https://www.primeintellect.ai/blog/synthetic-2"
        },
        {
          "kind": "public",
          "quote": "By aggregating data from publicly available datasets, using existing research repositories,",
          "url": "https://www.primeintellect.ai/blog/synthetic-2"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "prolific",
      "name": "Prolific",
      "website": "https://prolific.co",
      "what": "Supplies human data, evaluations and human feedback for AI developers and labs, via API or managed services.",
      "says": {
        "quote": "Run rigorous evaluations with verified people and domain experts.",
        "url": "https://www.prolific.com/ai"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Measure capability, safety, and quality with human evaluations, SME verification, and rubric design.",
          "url": "https://www.prolific.com/ai"
        },
        {
          "kind": "human-data",
          "quote": "Generate preference data, human feedback, and judgment signals to align models with intended outcomes.",
          "url": "https://www.prolific.com/ai"
        },
        {
          "kind": "datasets",
          "quote": "End-to-end evaluation and training data. Managed for you From evaluator specification to data delivery - with the reproducibility and transparency your research demands.",
          "url": "https://www.prolific.com/ai-services"
        },
        {
          "kind": "annotation",
          "quote": "Alternatively, you can also use AI Task Builder for text annotations and evaluations.",
          "url": "https://www.prolific.com/ai"
        }
      ],
      "areas": [
        "coding",
        "medical",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Verified domain experts generating SFT data, instruction-response pairs, and specialist annotations across 80+ languages.",
          "url": "https://www.prolific.com/ai"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-16"
    },
    {
      "id": "proximal",
      "name": "Proximal",
      "website": "https://proximal.ai",
      "what": "Builds coding data and benchmarks, such as FrontierSWE, for training and evaluating frontier coding agents.",
      "says": {
        "quote": "We build the data engine behind the next generation of autonomous coding agents.",
        "url": "https://proximal.ai"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "FrontierSWE is an effort to test coding agents on the hardest ultra-long horizon technical challenges.",
          "url": "https://www.proximal.ai/blog/frontierswe"
        },
        {
          "kind": "datasets",
          "quote": "Proximal is a new data company.",
          "url": "https://proximal.ai"
        }
      ],
      "areas": [
        "coding",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "We use synthetic data in many different parts of our data generation and want to scale our methods to generate entire codebases from scratch.",
          "url": "https://www.proximal.ai/blog/proximal"
        },
        {
          "kind": "experts",
          "quote": "to curate problems that experts outside of Proximal are uniquely aware of.",
          "url": "https://www.proximal.ai/blog/frontierswe"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, USA",
        "quote": "© 2026 Proximal, San Francisco, CA",
        "url": "https://proximal.ai"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "quesma",
      "name": "Quesma",
      "website": "https://quesma.com",
      "what": "Provides open-source coding-agent benchmarks and training environments, and cost analytics for coding-agent sessions.",
      "says": {
        "quote": "Quesma turns Claude Code, Codex, and Cursor sessions into a clear view of what the agents did, what it cost, and what you got back.",
        "url": "http://quesma.io/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We built training environments for difficult, multi-hour agent tasks.",
          "url": "http://quesma.io/"
        },
        {
          "kind": "evals",
          "quote": "Open-source benchmarks for evaluating AI coding agents on real-world software engineering tasks.",
          "url": "https://quesma.com/benchmarks"
        }
      ],
      "areas": [
        "coding",
        "security"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "reasoncore",
      "name": "ReasonCore",
      "website": "https://reasoncore.ai/",
      "what": "Sells financial fraud benchmarks and RL- and SFT-ready training data packs for frontier AI models.",
      "says": {
        "quote": "RL- and SFT-ready training pack for financial fraud",
        "url": "https://benchmarks.reasoncore.ai/financial-fraud"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "We evaluated four frontier LLMs on financial fraud risk classification across 30 expert-curated scenarios spanning Account Takeover and Money Mule typologies.",
          "url": "https://benchmarks.reasoncore.ai/financial-fraud"
        },
        {
          "kind": "datasets",
          "quote": "RL- and SFT-ready training pack for financial fraud",
          "url": "https://benchmarks.reasoncore.ai/financial-fraud"
        },
        {
          "kind": "human-data",
          "quote": "Every scenario is independently authored by a domain expert, passes through two levels of expert review and editorial compilation, and gets Pass@8 sampled to confirm adversarial validity and quality.",
          "url": "https://benchmarks.reasoncore.ai/financial-fraud"
        },
        {
          "kind": "annotation",
          "quote": "Content, labeling, and adjudication are led by former executives in Payments, AML, and risk operations",
          "url": "https://benchmarks.reasoncore.ai/financial-fraud"
        }
      ],
      "areas": [
        "finance"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Every scenario is independently authored by a domain expert, passes through two levels of expert review and editorial compilation, and gets Pass@8 sampled to confirm adversarial validity and quality.",
          "url": "https://benchmarks.reasoncore.ai/financial-fraud"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "refresh",
      "name": "Refresh",
      "website": "https://refresh.dev",
      "what": "Builds simulated worlds and evals for coding and computer use, which frontier AI labs use to train and benchmark their models.",
      "says": {
        "quote": "Refresh builds simulated worlds for coding and computer use, with verifiable rewards that move models forward.",
        "url": "https://refresh.dev"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We help AI labs achieve SOTA performance on tool-using verticals through high-quality, hand-crafted environments that expose capability gaps.",
          "url": "https://refresh.dev/one-pager"
        },
        {
          "kind": "evals",
          "quote": "Labs use Refresh environments and evals to measure and push the frontier of what their models can do",
          "url": "https://refresh.dev"
        },
        {
          "kind": "datasets",
          "quote": "An SFT dataset of real developer workflows from 2K repositories with pre-resolved dependencies and Docker containers.",
          "url": "https://refresh.dev/one-pager"
        },
        {
          "kind": "human-data",
          "quote": "Refresh builds environments by partnering with expert engineers in our network to source real tasks that break the leading models.",
          "url": "https://refresh.dev"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Refresh builds environments by partnering with expert engineers in our network to source real tasks that break the leading models.",
          "url": "https://refresh.dev"
        },
        {
          "kind": "synthetic",
          "quote": "Scaling simulation engines in general is human capital intensive, using a weakly supervised approach we can artificially generate and validate hundreds and thousands of environments on demand.",
          "url": "https://refresh.dev/one-pager"
        },
        {
          "kind": "replica",
          "quote": "High-fidelity clones of full software suites, wiring together multiple connected applications, from EHRs to enterprise apps and beyond.",
          "url": "https://refresh.dev"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2025 by Erik Quintanilla and Christopher Settles, Refresh has 8 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/refresh"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Erik Quintanilla and Christopher Settles, Refresh has 8 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/refresh"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "revelo",
      "name": "Revelo",
      "website": "https://revelo.com",
      "what": "Sells expert-authored code datasets, RL environments and evals to AI labs, and lets companies hire software engineers in Latin America.",
      "says": {
        "quote": "The data lab specializing in code: expert-authored data, RL environments, and evals for AI labs, plus elite software engineers to hire in Latin America.",
        "url": "https://revelo.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Expert-built, containerized tasks for long-horizon agent work, with verifiers, proof of solvability, and pass rates.",
          "url": "https://www.revelo.com/human-data-for-llm-training"
        },
        {
          "kind": "datasets",
          "quote": "Pre-built, expert-authored datasets and RL environments with tasks, verifiers, and trajectories, ready to license.",
          "url": "https://www.revelo.com/human-data-for-llm-training"
        },
        {
          "kind": "human-data",
          "quote": "Demonstrations authored or improved by experts, with chain-of-thought, diffs, and trajectories where you need them.",
          "url": "https://www.revelo.com/human-data-for-llm-training"
        },
        {
          "kind": "evals",
          "quote": "Domain experts judge outputs against task-specific rubrics.",
          "url": "https://www.revelo.com/human-data-for-llm-training"
        }
      ],
      "areas": [
        "coding",
        "games",
        "science",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Today, those same engineers author the datasets, RL environments, and evals that leading AI labs use to train and evaluate their models.",
          "url": "https://www.revelo.com/human-data-for-llm-training"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Miami, United States",
        "quote": "Today, Revelo is a series B, hyper growth tech startup headquartered in Miami with over 250 remote employees distributed across United States and Latin America.",
        "url": "https://www.revelo.com/about-us"
      },
      "founded": {
        "value": 2015,
        "quote": "Lachlan and Lucas set out to bridge the gap by founding Revelo in 2015.",
        "url": "https://www.revelo.com/about-us"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "rise-data-labs",
      "name": "Rise Data Labs",
      "website": "https://risedatalabs.com",
      "what": "Provides human expert data labeling, model evaluation, and RL environments to companies building AI models.",
      "says": {
        "quote": "We power training, evaluation, and next-gen generative AI with elite US-based, context-aware human trainers paired with automation that scales quality.",
        "url": "https://risedatalabs.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Custom RL environments built from your real workflows train agents that perform in production, not just in tests.",
          "url": "https://risedatalabs.com/solutions/rl-environment"
        },
        {
          "kind": "evals",
          "quote": "Comprehensive model evaluation and benchmarking services.",
          "url": "https://risedatalabs.com/solutions/model-evaluation-benchmarking"
        },
        {
          "kind": "human-data",
          "quote": "Training AI models needs high quality human feedback to produce high quality output.",
          "url": "https://risedatalabs.com"
        },
        {
          "kind": "datasets",
          "quote": "High-fidelity AI training data combining expert human annotation with synthetic generation.",
          "url": "https://risedatalabs.com/solutions/synthetic-human-data-pipelines"
        },
        {
          "kind": "annotation",
          "quote": "Expert labeling across text, code, and multimodal data with domain-grade quality control.",
          "url": "https://risedatalabs.com/solutions/synthetic-human-data-pipelines"
        }
      ],
      "areas": [
        "enterprise",
        "finance",
        "legal",
        "games",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Domain-matched US professionals delivering context-aware, high-accuracy labeled data.",
          "url": "https://risedatalabs.com/solutions/synthetic-human-data-pipelines"
        },
        {
          "kind": "customer",
          "quote": "We synthesize your first-party data into custom RL environments that reflect real goals, constraints, and outcomes closing the gap between simulation and production.",
          "url": "https://risedatalabs.com/solutions/rl-environment"
        },
        {
          "kind": "synthetic",
          "quote": "Fill data gaps and accelerate coverage with synthetic pipelines calibrated to your model.",
          "url": "https://risedatalabs.com/solutions/synthetic-human-data-pipelines"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "rl-labs",
      "name": "RL Labs",
      "website": "https://rllabs.ai",
      "what": "Builds long-horizon RL environments for training frontier AI models.",
      "says": {
        "quote": "We build long-horizon RL environments for training frontier models",
        "url": "https://rllabs.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We build long-horizon RL environments for training frontier models",
          "url": "https://rllabs.ai"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "security"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "runloop",
      "name": "Runloop",
      "website": "https://runloop.ai/",
      "what": "Provides secure code sandboxes, agent evaluations and RL data generation for teams building AI agents.",
      "says": {
        "quote": "Launch AI agents on secure code sandboxes, refine with evaluations, and ship on AI infrastructure built for enterprise scale",
        "url": "https://runloop.ai/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Evaluate models against your real workload, not mock benchmarks",
          "url": "http://runloop.ai/"
        },
        {
          "kind": "datasets",
          "quote": "Runloop gives post-training teams one execution layer for three workloads: RL rollouts, SFT data generation, and verified trajectory generation. Each rollout runs in an isolated sandbox provisioned from a reproducible Blueprint, executes against production-like dependencies, and returns structured outcomes your training pipeline can consume.",
          "url": "https://runloop.ai/use-case/rl-fine-tuning"
        },
        {
          "kind": "rl-environments",
          "quote": "Runloop gives post-training teams one execution layer for three workloads: RL rollouts, SFT data generation, and verified trajectory generation. Each rollout runs in an isolated sandbox provisioned from a reproducible Blueprint, executes against production-like dependencies, and returns structured outcomes your training pipeline can consume.",
          "url": "https://runloop.ai/use-case/rl-fine-tuning"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-16"
    },
    {
      "id": "sama",
      "name": "Sama",
      "website": "https://sama.com",
      "what": "Sells data annotation, validation and model evaluation services to companies building AI models.",
      "says": {
        "quote": "Sama delivers human-verified annotation, validation, and evaluation at scale",
        "url": "https://www.sama.com/services"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Sama combines automation with expert human-verified data for computer vision, NLP, and multimodal AI",
          "url": "https://www.sama.com/services"
        },
        {
          "kind": "annotation",
          "quote": "Expert-led labeling for vision, multimodal, and sensor data",
          "url": "https://www.sama.com/services"
        },
        {
          "kind": "evals",
          "quote": "Score real-world performance to guide retraining, iteration, and confident deployment.",
          "url": "https://www.sama.com/services"
        }
      ],
      "areas": [
        "vision",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Sama delivers expert-led, HITL annotation, validation, and evaluation services that improve model accuracy, reduce rework, and support predictable scaling across the AI lifecycle.",
          "url": "https://www.sama.com/services"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2008,
        "quote": "She founded Sama in 2008 driven by the belief that “talent is equally distributed, but opportunity is not.”",
        "url": "https://www.sama.com/about"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "sapien",
      "name": "Sapien",
      "website": "https://sapien.io",
      "what": "Offers Proof of Quality, a protocol in which a panel of experts scores AI data or agent outputs and issues signed, checkable reports to AI teams.",
      "says": {
        "quote": "A panel of independent experts scores each item against your rubric, and the result is sealed as a permanent, auditable record.",
        "url": "http://docs.sapien.io/how-it-works"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Check your model's answers and your agent's steps against your rules, using a group of reviewers instead of a single grader.",
          "url": "http://docs.sapien.io/how-it-works"
        },
        {
          "kind": "annotation",
          "quote": "When an expert labels a scan, an auditor flags a vulnerability, or a model makes a call, PoQ answers the question every AI team asks",
          "url": "http://docs.sapien.io/how-it-works"
        }
      ],
      "areas": [
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "A panel of independent experts scores each item against your rubric",
          "url": "http://docs.sapien.io/how-it-works"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "scale",
      "name": "Scale AI",
      "website": "https://scale.com",
      "what": "Provides training data, RL environments, and model evaluations to AI labs, governments, and enterprises.",
      "says": {
        "quote": "Scale delivers proven data, evaluations, and outcomes to AI labs, governments, and the Fortune 500.",
        "url": "https://scale.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Scale AI provides RL environments for training agents on desktop VMs, MCP servers, and workflows.",
          "url": "https://scale.com/rlenvironments"
        },
        {
          "kind": "datasets",
          "quote": "High-quality training data, annotations, and RLHF to power the world's most advanced AI models.",
          "url": "http://scale.com/about"
        },
        {
          "kind": "human-data",
          "quote": "We source contributors with precision (25% have advanced degrees) and deliver at the bar frontier AI demands.",
          "url": "https://scale.com"
        },
        {
          "kind": "evals",
          "quote": "Rigorous model evaluations and red-teaming to measure, benchmark, and improve AI performance.",
          "url": "http://scale.com/about"
        },
        {
          "kind": "annotation",
          "quote": "Access customized data annotations, model evaluations, and RLHF data via API, SDK, or web frontend.",
          "url": "https://api-reference.scale.com/docs/genai-data-engine"
        },
        {
          "kind": "simulation",
          "quote": "Train AI agents in realistic simulated worlds.",
          "url": "https://scale.com/blog/rl-environments"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "robotics",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Domain experts curate artifacts that capture the complexity, ambiguity, and edge cases of real professional work.",
          "url": "https://scale.com/blog/rl-environments"
        },
        {
          "kind": "replica",
          "quote": "They mirror real computer and API-based systems, supporting rich logs and application state that can be used for programmatic and rubric-based evaluation.",
          "url": "https://scale.com/rlenvironments"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2016,
        "quote": "Since 2016, we've been at the forefront; from autonomous vehicles to frontier models solving the world's hardest problems.",
        "url": "https://scale.com"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "sepal-ai",
      "name": "Sepal AI",
      "website": "https://www.sepalai.com",
      "what": "Produces training data, evaluation benchmarks and RL environments for AI labs and enterprises on its data development platform.",
      "says": {
        "quote": "We’ve built Sepal AI - the data development platform that enables you to curate useful datasets.",
        "url": "https://www.ycombinator.com/companies/sepal-ai"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We’ve built Sepal AI - the data development platform that enables you to curate useful datasets.",
          "url": "https://www.ycombinator.com/companies/sepal-ai"
        },
        {
          "kind": "human-data",
          "quote": "We bring data generation tooling, human experts, synthetic data augmentation, and rigorous quality control into one platform so you can manage the production of high-quality datasets.",
          "url": "https://www.ycombinator.com/companies/sepal-ai"
        },
        {
          "kind": "evals",
          "quote": "Cell and Molecular Biology Benchmark: An original benchmark to evaluate complex reasoning across models.",
          "url": "https://www.ycombinator.com/companies/sepal-ai"
        }
      ],
      "areas": [
        "finance",
        "medical",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We’ve built a network of 20k+ experts across STEM and professional services",
          "url": "https://www.ycombinator.com/companies/sepal-ai"
        },
        {
          "kind": "synthetic",
          "quote": "We bring data generation tooling, human experts, synthetic data augmentation, and rigorous quality control into one platform",
          "url": "https://www.ycombinator.com/companies/sepal-ai"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2024 by Kat Hu and Robi Lin, Sepal AI has 15 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/sepal-ai"
      },
      "founded": {
        "value": 2024,
        "quote": "Sepal AI was founded in 2024 by engineers and operators from Vercel and Turing.",
        "url": "https://www.ycombinator.com/companies/sepal-ai"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "sieve",
      "name": "Sieve",
      "website": "https://sieve.ai/",
      "what": "Sells multimodal video, audio, image and interaction datasets, evaluation sets and environments to frontier AI labs for training.",
      "says": {
        "quote": "Sieve builds the data and environments frontier AI labs use to train the next generation of multimodal systems.",
        "url": "https://www.sieve.ai/about"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Sieve builds the data and environments frontier AI labs use to train the next generation of multimodal systems.",
          "url": "https://www.sieve.ai/about"
        },
        {
          "kind": "datasets",
          "quote": "We package training-ready datasets, evaluation sets, and environments for secure delivery.",
          "url": "https://sieve.ai/"
        },
        {
          "kind": "evals",
          "quote": "We package training-ready datasets, evaluation sets, and environments for secure delivery.",
          "url": "https://sieve.ai/"
        },
        {
          "kind": "annotation",
          "quote": "We add dense labels, pairings, temporal alignment, transcripts, action metadata, and human QA at scale.",
          "url": "https://sieve.ai/"
        }
      ],
      "areas": [
        "computer-use",
        "robotics",
        "vision",
        "speech",
        "creative"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "We capture and aggregate multimodal data across real-world, digital, and simulated environments.",
          "url": "https://sieve.ai/"
        },
        {
          "kind": "licensed",
          "quote": "Earn revenue by licensing content you already own, or by recording the skilled work your team already does.",
          "url": "https://www.sieve.ai/partners"
        }
      ],
      "buysData": {
        "quote": "Earn revenue by licensing content you already own, or by recording the skilled work your team already does.",
        "url": "https://www.sieve.ai/partners"
      },
      "hq": null,
      "founded": {
        "value": 2022,
        "quote": "Abhi and I started the company in 2022 building developer tools for computer vision.",
        "url": "https://www.sieve.ai/blog/reintro"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "simpleclosure",
      "name": "SimpleClosure",
      "website": "https://simpleclosure.com",
      "what": "Runs business dissolutions for founders and sells their startups' source code and workspace data to AI buyers.",
      "says": {
        "quote": "AssetHub helps startups monetize their assets, like SourceCode or WorkspaceData to vetted AI buyers looking for real-world work product.",
        "url": "https://simpleclosure.com/product/assethub"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "We work with a vetted network of strategic buyers, including leading AI foundation labs, reinforcement learning environment builders, and other technology companies looking for real-world data for AI training.",
          "url": "https://simpleclosure.com/product/assethub"
        }
      ],
      "areas": [
        "coding",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Founders working with SimpleClosure can recover it through two distinct ways: the codebases they built, and the workspace data from how their teams actually worked.",
          "url": "https://simpleclosure.com/product/assethub"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2023,
        "quote": "SimpleClosure was founded in 2023 to address a gap in how startup closure is handled.",
        "url": "https://simpleclosure.com/about"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "snorkel",
      "name": "Snorkel AI",
      "website": "https://snorkel.ai/research/",
      "what": "Develops specialized training data, benchmarks, and environments for frontier AI labs and enterprise AI teams.",
      "says": {
        "quote": "Snorkel helps frontier labs and AI teams develop specialized training data and environments that set their models and agents apart.",
        "url": "https://snorkel.ai/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "The Snorkel Data Series is a set of research-defined, non-exclusive dataset and environment products, refreshed quarterly as models advance.",
          "url": "https://snorkel.ai/frequently-asked-questions"
        },
        {
          "kind": "rl-environments",
          "quote": "Whether it’s a benchmark, an eval set, RL environments, or specialized training data, we can help on essentially any data or evaluation project in a high-stakes domain.",
          "url": "https://snorkel.ai/frequently-asked-questions"
        },
        {
          "kind": "human-data",
          "quote": "We deliver expert data development as a service: datasets, benchmarks, evaluations, and environments built with our experts, rather than a SaaS app your team runs.",
          "url": "https://snorkel.ai/frequently-asked-questions"
        },
        {
          "kind": "evals",
          "quote": "Frontier teams come to us for benchmark creation and RL/agentic evaluation.",
          "url": "https://snorkel.ai/frequently-asked-questions"
        },
        {
          "kind": "annotation",
          "quote": "Learn how Snorkel’s Data-as-a-Service helps teams label, refine, and evaluate high-quality, domain-specific datasets for your projects.",
          "url": "https://snorkel.ai/company"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "legal",
        "science"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We deliver expert data development as a service: datasets, benchmarks, evaluations, and environments built with our experts, rather than a SaaS app your team runs.",
          "url": "https://snorkel.ai/frequently-asked-questions"
        },
        {
          "kind": "customer",
          "quote": "Enterprises come to build reliable, domain-specific AI grounded in their own data and operating knowledge.",
          "url": "https://snorkel.ai/frequently-asked-questions"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2019,
        "quote": "Founded out of the Stanford AI Lab in 2019.",
        "url": "https://snorkel.ai/company"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "superannotate",
      "name": "SuperAnnotate",
      "website": "https://superannotate.com",
      "what": "Provides expert human data, annotation, evaluation, and RL environment services to companies building AI models.",
      "says": {
        "quote": "Give your models the expert human data they need – and simplify your AI development today.",
        "url": "http://superannotate.com/ai-data-services"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Preference data and supervised fine-tuning sets built to your guidelines.",
          "url": "https://www.superannotate.com/solutions-enterprise"
        },
        {
          "kind": "rl-environments",
          "quote": "SuperAnnotate helps build reliable RL environments with realistic tasks and rich reward systems",
          "url": "https://www.superannotate.com/blog/rl-environments"
        },
        {
          "kind": "evals",
          "quote": "Benchmark and stress-test models with expert reviewers and structured rubrics.",
          "url": "https://www.superannotate.com/solutions-enterprise"
        },
        {
          "kind": "annotation",
          "quote": "Text, image, video, and audio labeled in one place – at frontier scale.",
          "url": "https://www.superannotate.com/solutions-enterprise"
        },
        {
          "kind": "datasets",
          "quote": "Versioned, audited datasets with full lineage from raw signal to train set.",
          "url": "https://www.superannotate.com/solutions-enterprise"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "math",
        "robotics",
        "speech",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Physicians, attorneys, engineers and scientists help train frontier models with us",
          "url": "http://superannotate.com/ai-data-services"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "surge",
      "name": "Surge AI",
      "website": "https://surgehq.ai/products",
      "what": "Provides expert-built training data, RL environments, and evaluations to AI labs and enterprises building AI models.",
      "says": {
        "quote": "Training data, evals, and RL environments built by the same experts behind our frontier-lab work.",
        "url": "https://surgehq.ai/ots"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We create rich, complex RL environments that challenge agentic models in novel ways, and design verifiers that reward their behavior.",
          "url": "https://www.surgehq.ai/products"
        },
        {
          "kind": "datasets",
          "quote": "Our off-the-shelf datasets are the same: thousands of hours of expert reasoning, pre-built and ready to use today, spanning RL environments, coding, and the core capabilities every model needs.",
          "url": "https://www.surgehq.ai/products"
        },
        {
          "kind": "human-data",
          "quote": "We bootstrap model capabilities through demonstrations: teaching them to use computers, navigate the web, and reason for the first time.",
          "url": "https://www.surgehq.ai/products"
        },
        {
          "kind": "evals",
          "quote": "Test models and AI systems against custom benchmarks and evaluation sets before they reach customers.",
          "url": "https://surgehq.ai/enterprise"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "vision",
        "speech",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Training data, evals, and RL environments built by the same experts behind our frontier-lab work.",
          "url": "https://surgehq.ai/ots"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "symbal",
      "name": "Symbal",
      "website": "https://www.symbal.ai",
      "what": "Develops evals, benchmarks, and post-training datasets for frontier AI labs, and deploys enterprise AI agents and automations.",
      "says": {
        "quote": "We develop evals, benchmarks, and post-training datasets to advance multimodal model performance.",
        "url": "https://www.symbal.ai/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Private evals, and data for SWE, Tau, MLE, MMMU, and more",
          "url": "https://www.symbal.ai/"
        },
        {
          "kind": "datasets",
          "quote": "from building high-quality training datasets to delivering production-ready models",
          "url": "https://www.symbal.ai/"
        },
        {
          "kind": "annotation",
          "quote": "VLA annotations, tele-op data, and sims",
          "url": "https://www.symbal.ai/"
        },
        {
          "kind": "simulation",
          "quote": "VLA annotations, tele-op data, and sims",
          "url": "https://www.symbal.ai/"
        }
      ],
      "areas": [
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "math",
        "speech",
        "vision",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Symbal partners with enterprises, converting your first-party data into a live digital map of your organization's processes.",
          "url": "https://www.symbal.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "synthetic-sciences",
      "name": "Synthetic Sciences",
      "website": "https://syntheticsciences.ai",
      "what": "Builds AI co-scientist tools for research teams, plus RL environments and evaluations for AI labs training models on research.",
      "says": {
        "quote": "An AI research lab building the tools, evaluations and environments that enable models to accelerate discovery in the empirical sciences.",
        "url": "https://syntheticsciences.ai/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We're building LHRL environments to post-train models on ML research workflows.",
          "url": "https://www.ycombinator.com/companies/synthetic-sciences"
        },
        {
          "kind": "evals",
          "quote": "we build long-horizon evaluations of whether models can form hypotheses",
          "url": "https://syntheticsciences.ai/"
        }
      ],
      "areas": [
        "science"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Every correction, accept/reject, trace, and outcome becomes training data.",
          "url": "https://www.ycombinator.com/companies/synthetic-sciences"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Founded in 2025 by Aayam Bansal and Ishaan Gangwani, Synthetic Sciences has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/synthetic-sciences"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Aayam Bansal and Ishaan Gangwani, Synthetic Sciences has 2 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/synthetic-sciences"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "tacit-labs",
      "name": "Tacit Labs",
      "website": "https://tacitlabs.co",
      "what": "Builds data, evaluations, and research environments for frontier AI models in biology, chemistry, and drug development.",
      "says": {
        "quote": "Tacit Labs is building data, evaluations, and research environments unlocking scientific discovery capabilities for frontier models.",
        "url": "https://www.tacitlabs.co/join-us"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Tacit is starting by building realistic, long-horizon evaluations that test whether AI systems can make",
          "url": "https://www.tacitlabs.co/introducing-tacit-labs"
        },
        {
          "kind": "rl-environments",
          "quote": "Help define what makes an environment useful for training, evaluation, and model improvement.",
          "url": "https://www.tacitlabs.co/join-us"
        },
        {
          "kind": "datasets",
          "quote": "Tacit Labs is building data, evaluations, and research environments unlocking scientific discovery capabilities for frontier models.",
          "url": "https://www.tacitlabs.co/join-us"
        }
      ],
      "areas": [
        "science"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "augmented with real-world experimental data and contextual",
          "url": "https://www.tacitlabs.co/introducing-tacit-labs"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "taskus",
      "name": "TaskUs",
      "website": "https://www.taskus.com/",
      "what": "Provides data annotation and training data services for companies building AI product, including autonomous vehicle and robotics systems.",
      "says": {
        "quote": "TaskUs and V7 are combining their expertise in AI and data annotation to optimize training data creation for AI product development at some of the world’s biggest technology companies.",
        "url": "https://www.taskus.com/press-release/taskus-and-v7-announce-strategic-partnership-to-enable-enterprise-ai-product-delivery/"
      },
      "sells": [
        {
          "kind": "annotation",
          "quote": "TaskUs and V7 are combining their expertise in AI and data annotation to optimize training data creation for AI product development at some of the world’s biggest technology companies.",
          "url": "https://www.taskus.com/press-release/taskus-and-v7-announce-strategic-partnership-to-enable-enterprise-ai-product-delivery/"
        },
        {
          "kind": "human-data",
          "quote": "already proving successful in enterprise healthcare projects.",
          "url": "https://www.taskus.com/press-release/taskus-and-v7-announce-strategic-partnership-to-enable-enterprise-ai-product-delivery/"
        },
        {
          "kind": "datasets",
          "quote": "We helped a leading tech giant to build a diverse, egocentric dataset to solve this.",
          "url": "https://www.taskus.com/insights/scaling-global-ground-truth-for-consumer-robotics/"
        }
      ],
      "areas": [
        "enterprise",
        "medical",
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "TaskUs is training its teams of subject matter experts to accelerate training data delivery using V7’s model-assisted labeling stack.",
          "url": "https://www.taskus.com/press-release/taskus-and-v7-announce-strategic-partnership-to-enable-enterprise-ai-product-delivery/"
        },
        {
          "kind": "captured",
          "quote": "Participants recorded their spaces in 1080p from a first-person point of view to ensure real-world perspective.",
          "url": "https://www.taskus.com/insights/scaling-global-ground-truth-for-consumer-robotics/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "taste-labs",
      "name": "Taste Labs",
      "website": "https://tastelabs.com",
      "what": "Runs a human data network of designers and annotation infrastructure for AI labs, tech companies and brands.",
      "says": {
        "quote": "TasteMakers is a human data and creative network working with leading AI labs, tech companies, and brands.",
        "url": "https://tastemakers.tastelabs.com/"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "TasteMakers is a human data and creative network working with leading AI labs, tech companies, and brands.",
          "url": "https://tastemakers.tastelabs.com/"
        }
      ],
      "areas": [
        "creative"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "TasteMakers is a curated collective of designers who share a high bar for quality, not a volume marketplace.",
          "url": "https://tastemakers.tastelabs.com/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "telus-digital",
      "name": "TELUS Digital",
      "website": "https://www.telusdigital.com/",
      "what": "Provides customer experience services, digital solutions and AI training data to global brands.",
      "says": {
        "quote": "Discover TELUS Digital - a customer experience innovator that designs, builds & delivers next-gen digital solutions for global and disruptive brands.",
        "url": "https://www.telusdigital.com/"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Frontier AI training data, engineering and intelligent automation to accelerate responsible AI rooted in real ROI.",
          "url": "https://www.telusdigital.com/"
        },
        {
          "kind": "annotation",
          "quote": "When we look for partners to label data, we look for groups that have a wellness-first perspective and the capability to bring rigor to a complex problem.",
          "url": "https://www.telusdigital.com/"
        }
      ],
      "areas": [
        "enterprise",
        "finance",
        "robotics",
        "security"
      ],
      "madeFrom": [],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "general-data",
      "name": "The General Data Company",
      "website": "https://thegeneraldata.com",
      "what": "Builds training data, evaluations and RL environments from industrial companies' operational records, for AI labs and for those companies.",
      "says": {
        "quote": "We build training data, evaluations and RL environments from the real operational records of industrial companies, for AI labs and for the companies themselves.",
        "url": "https://thegeneraldata.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "The General Data Company builds reinforcement learning environments for AI labs from real industrial workflows",
          "url": "https://thegeneraldata.com/rl-environments"
        },
        {
          "kind": "datasets",
          "quote": "Decades of real operational data that never reached the internet, from drawings and change orders to quality reports, service histories and tenders.",
          "url": "https://thegeneraldata.com"
        },
        {
          "kind": "evals",
          "quote": "Evaluation sets are held-out, never-published document tasks for benchmarking extraction, reasoning and agent workflows.",
          "url": "https://thegeneraldata.com/data-licensing"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "From companies that agreed to it in writing.",
          "url": "https://thegeneraldata.com/faq"
        },
        {
          "kind": "customer",
          "quote": "A model trained on your company’s own history, running in your own cloud.",
          "url": "https://thegeneraldata.com"
        }
      ],
      "buysData": {
        "quote": "Companies license records they no longer need through our data partnerships",
        "url": "https://thegeneraldata.com/faq"
      },
      "hq": {
        "value": "San Francisco, USA",
        "quote": "The General Data Company is an AI training-data company. We’re based in San Francisco.",
        "url": "https://thegeneraldata.com/faq"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "theta",
      "name": "Theta",
      "website": "https://thetasoftware.com",
      "what": "Builds simulation environments that train AI agents on business workflows for frontier labs and enterprises.",
      "says": {
        "quote": "Theta translates real-world workspaces into simulation environments, training expert-level agents that deeply understand your work.",
        "url": "https://thetasoftware.com"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Environments are the digital workspaces where agents learn to complete long-horizon tasks with autonomy and reliability.",
          "url": "https://thetasoftware.com"
        },
        {
          "kind": "evals",
          "quote": "We're excited to announce CUB, a first-of-its-kind benchmark for computer and browser use agents.",
          "url": "https://thetasoftware.com/blog/introducing-cub"
        },
        {
          "kind": "human-data",
          "quote": "We synthesize your first party data and our human expert data into environments that reflect the real goals, workflows, and outcomes of your business.",
          "url": "https://thetasoftware.com"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "medical"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "We synthesize your first party data and our human expert data into environments that reflect the real goals, workflows, and outcomes of your business.",
          "url": "https://thetasoftware.com"
        },
        {
          "kind": "experts",
          "quote": "We've worked with accountants, investment bankers, doctors, and other experts to create high-quality domain-specific tasks.",
          "url": "https://thetasoftware.com/blog/introducing-cub"
        },
        {
          "kind": "synthetic",
          "quote": "These tasks involve real-world workflows and software tools across these industries, with synthetic versions of enterprise platforms such as SAP and CapIQ.",
          "url": "https://thetasoftware.com/blog/introducing-cub"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Specialized AI for Every Job. Founded in 2025 by Rayan Garg, Tanmay Sharma, and Gurvir Singh, Theta has 7 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/theta"
      },
      "founded": {
        "value": 2025,
        "quote": "Specialized AI for Every Job. Founded in 2025 by Rayan Garg, Tanmay Sharma, and Gurvir Singh, Theta has 7 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/theta"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "theta-lab",
      "name": "Theta Lab",
      "website": "https://thetalab.tech",
      "what": "Captures production AI agent runs, rebuilds hard cases as RL environments, and trains custom models for teams shipping agents.",
      "says": {
        "quote": "thetalab trains a custom model on your RL environment, so proven agent behavior becomes cheaper, more consistent, and yours to control, not a generic API call you rent.",
        "url": "https://thetalab.tech/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Rebuild the hard cases as RL environments to train in.",
          "url": "https://thetalab.tech/"
        },
        {
          "kind": "evals",
          "quote": "It captures live agent executions, makes failures replayable, and gives you the data for debugging, evals, and retraining.",
          "url": "https://thetalab.tech/"
        }
      ],
      "areas": [
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "customer",
          "quote": "Deployed on-prem · trained on your data",
          "url": "https://thetalab.tech/"
        },
        {
          "kind": "replica",
          "quote": "We don't build simplified mocks",
          "url": "https://www.thetalab.tech/blog/why-rl-environments-matter"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "toloka",
      "name": "Toloka",
      "website": "https://toloka.ai",
      "what": "Provides curated data, RL environments and evaluations, made with human experts, for AI agents and models.",
      "says": {
        "quote": "Toloka is a provider of expertly curated data for AI agents and models development.",
        "url": "http://toloka.ai/about-us"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Toloka builds environments & RL-gyms, collecting trajectories and graded eval signals to train and evaluate AI agents.",
          "url": "https://toloka.ai/agent-data"
        },
        {
          "kind": "datasets",
          "quote": "Our expertly curated datasets ensure precision and reliability, enabling superior fine-tuning and performance for your models.",
          "url": "https://toloka.ai/datasets"
        },
        {
          "kind": "human-data",
          "quote": "Agent trajectory demonstrations and step-by-step evaluations across tool-use workflows",
          "url": "https://toloka.ai"
        },
        {
          "kind": "evals",
          "quote": "Assessing agent performance and identifying vulnerabilities",
          "url": "https://toloka.ai"
        },
        {
          "kind": "annotation",
          "quote": "Toloka has developed a state-of-the-art technology platform for data labeling and has over 10 years of managing human efforts, ensuring operational excellence at scale.",
          "url": "http://toloka.ai/about-us"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "math",
        "security",
        "robotics",
        "vision",
        "creative",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Now, Toloka collaborates with data workers from 100+ countries speaking 40+ languages across, 50+ knowledge domains and 120+ subdomains.",
          "url": "http://toloka.ai/about-us"
        },
        {
          "kind": "synthetic",
          "quote": "Generate synthetic data from scratch, or validate your pre-generated data at any stage.",
          "url": "https://toloka.ai/hybrid-data"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Amsterdam, Netherlands",
        "quote": "Toloka is a European company. Our global headquarters is located in Amsterdam.",
        "url": "http://toloka.ai/about-us"
      },
      "founded": {
        "value": 2014,
        "quote": "She established the company in 2014",
        "url": "http://toloka.ai/about-us"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "trajectory-labs",
      "name": "Trajectory Labs",
      "website": "https://trajectorylabs.com",
      "what": "Provides AI safety evaluations, red teaming, and RL environments to frontier AI labs.",
      "says": {
        "quote": "Trajectory Labs, PBC provides AI safety evaluations, red teaming, and RL environments to frontier labs.",
        "url": "https://www.trajectorylabs.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Trajectory Labs, PBC provides AI safety evaluations, red teaming, and RL environments to frontier labs.",
          "url": "https://www.trajectorylabs.com/"
        },
        {
          "kind": "evals",
          "quote": "Trajectory Labs, PBC provides AI safety evaluations, red teaming, and RL environments to frontier labs.",
          "url": "https://www.trajectorylabs.com/"
        }
      ],
      "areas": [
        "coding",
        "security"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Our expert red teamers continue to find new vulnerabilities in every frontier model we've tested and are relied on by frontier labs to keep their models safe.",
          "url": "https://www.trajectorylabs.com/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "traverse",
      "name": "Traverse",
      "website": "https://traverse.so",
      "what": "Runs a platform where industry professionals share expertise that helps train and improve AI models.",
      "says": {
        "quote": "Traverse Experts connects industry professionals with opportunities to share their expertise.",
        "url": "https://experts.traverse.so/"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Your knowledge helps train and improve AI models, shaping the future of artificial intelligence.",
          "url": "https://experts.traverse.so/"
        },
        {
          "kind": "rl-environments",
          "quote": "We capture real human behavior in highly controlled settings and turn judgment, taste, and long-horizon reasoning into high-fidelity environments for frontier models.",
          "url": "http://ycombinator.com/companies/traverse"
        },
        {
          "kind": "datasets",
          "quote": "Traverse captures what people actually think while doing their work; we are pioneering a new category of training data that is yet to exist.",
          "url": "http://ycombinator.com/companies/traverse"
        }
      ],
      "areas": [],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Traverse Experts connects industry professionals with opportunities to share their expertise.",
          "url": "https://experts.traverse.so/"
        },
        {
          "kind": "captured",
          "quote": "We capture real human behavior in highly controlled settings and turn judgment, taste, and long-horizon reasoning into high-fidelity environments for frontier models.",
          "url": "http://ycombinator.com/companies/traverse"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Lance Yan, Traverse has 1 employees based in San Francisco.",
        "url": "http://ycombinator.com/companies/traverse"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "troveo",
      "name": "Troveo",
      "website": "https://troveo.ai",
      "what": "Licenses non-public business, video, audio and gaming data, plus custom-sourced datasets and annotation, to frontier AI labs.",
      "says": {
        "quote": "We unlock non-public, real-world data for frontier AI",
        "url": "https://troveo.ai"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Curate from millions of hours of content and 100+ video, audio, and gaming datasets.",
          "url": "https://troveo.ai"
        },
        {
          "kind": "annotation",
          "quote": "Annotation pipelines designed for model training and evaluation at scale.",
          "url": "https://www.troveo.ai/sourcing"
        }
      ],
      "areas": [
        "vision",
        "speech",
        "robotics",
        "games",
        "enterprise",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "licensed",
          "quote": "Share a brief and we source data to your exact spec from our licensor network.",
          "url": "https://troveo.ai"
        },
        {
          "kind": "customer",
          "quote": "Troveo matches you with them, anonymizes your data to a documented standard, and licenses it on terms you approve",
          "url": "https://www.troveo.ai/business-data"
        },
        {
          "kind": "captured",
          "quote": "We activate exclusive supply networks to capture and deliver high-quality, model-aligned content at scale.",
          "url": "https://www.troveo.ai/sourcing"
        }
      ],
      "buysData": {
        "quote": "Share a brief and we source data to your exact spec from our licensor network.",
        "url": "https://troveo.ai"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "turing",
      "name": "Turing",
      "website": "https://www.turing.com/advance/rl-environments",
      "what": "Sells off-the-shelf and custom datasets, RL environments, and human experts to frontier AI labs in software engineering, enterprise knowledge work, and STEM.",
      "says": {
        "quote": "We offer frontier AI labs custom and off-the-shelf datasets, RL environments, and human experts in software engineering, enterprise knowledge work, and frontier STEM.",
        "url": "https://www.turing.com/frontier-ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Iterable UI and non-UI environments that agents can use to explore and learn long-horizon workflows.",
          "url": "https://www.turing.com/advance/rl-environments"
        },
        {
          "kind": "datasets",
          "quote": "We offer frontier AI labs custom and off-the-shelf datasets, RL environments, and human experts in software engineering, enterprise knowledge work, and frontier STEM.",
          "url": "https://www.turing.com/frontier-ai"
        },
        {
          "kind": "human-data",
          "quote": "We offer frontier AI labs custom and off-the-shelf datasets, RL environments, and human experts in software engineering, enterprise knowledge work, and frontier STEM.",
          "url": "https://www.turing.com/frontier-ai"
        },
        {
          "kind": "evals",
          "quote": "CEO Bench evaluates whether AI agents can complete long-horizon financial and operational work inside a complete, expert-authored fictional company.",
          "url": "https://www.turing.com/frontier-ai/enterprise-knowledge-work"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise",
        "finance",
        "legal",
        "medical",
        "science",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Research-grade and built by experts to hillclimb on key benchmarks.",
          "url": "https://www.turing.com/frontier-ai/enterprise-knowledge-work"
        },
        {
          "kind": "replica",
          "quote": "Interactive replicas of enterprise and consumer apps such as Jira, Salesforce, and Zendesk.",
          "url": "https://www.turing.com/advance/rl-environments"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco, United States",
        "quote": "© 2026 Turing Enterprises, Inc. All rights reserved. · 548 Market Street, PMB 18282, San Francisco, CA 94104",
        "url": "https://www.turing.com/frontier-ai/enterprise-knowledge-work"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "uber-ai-solutions",
      "name": "Uber AI Solutions",
      "website": "https://www.uber.com/us/en/ai-solutions/",
      "what": "Provides AI data annotation, collection, and evaluation services to enterprises, frontier labs, robotics firms, and autonomous vehicle developers.",
      "says": {
        "quote": "Uber AI Solutions partners with enterprises at every stage of the data lifecycle, offering human oversight, governed workflows, and seamless pipeline integration.",
        "url": "https://www.uber.com/us/en/ai-solutions/"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Tap into our global network of contributors and specialized experts to generate supervised fine-tuning (SFT) data and execute reinforcement learning and human feedback (RLHF) workflows.",
          "url": "https://www.uber.com/us/en/ai-solutions/frontier-labs/"
        },
        {
          "kind": "annotation",
          "quote": "Data annotation provides context and meaning to the data used to train generative and agentic AI models.",
          "url": "https://www.uber.com/us/en/ai-solutions/annotation/"
        },
        {
          "kind": "datasets",
          "quote": "We prepare ready-to-use and custom datasets supporting a range of AI use cases",
          "url": "https://www.uber.com/us/en/ai-solutions/annotation/"
        },
        {
          "kind": "evals",
          "quote": "Orchestrate automated and human-led evaluation pipelines to benchmark model performance, diagnose edge case failures, and guide systematic model refinement.",
          "url": "https://www.uber.com/us/en/ai-solutions/frontier-labs/"
        },
        {
          "kind": "rl-environments",
          "quote": "custom reinforced learning (RL) gyms built around your specific use case",
          "url": "https://www.uber.com/us/en/ai-solutions/gen-ai/"
        }
      ],
      "areas": [
        "enterprise",
        "robotics",
        "vision",
        "speech",
        "multilingual",
        "coding",
        "finance",
        "medical",
        "legal"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Tap into our global network of contributors and specialized experts to generate supervised fine-tuning (SFT) data and execute reinforcement learning and human feedback (RLHF) workflows.",
          "url": "https://www.uber.com/us/en/ai-solutions/frontier-labs/"
        },
        {
          "kind": "captured",
          "quote": "Our extensive global network of contributors and specialized experts gather real-world data, including point-of-interest (POI), audio, and video, through remote and targeted field tasks.",
          "url": "https://www.uber.com/us/en/ai-solutions/annotation/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "ulam",
      "name": "Ulam",
      "website": "https://www.ulam.ai",
      "what": "Builds reasoning datasets and evaluation systems for frontier AI labs and AI developers.",
      "says": {
        "quote": "Ulam builds research-grade reasoning datasets and evaluation systems for frontier AI labs and AI developers.",
        "url": "https://www.ulam.ai/"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Custom benchmark suites and eval protocols for labs that need private, difficult-to-overfit measures of frontier reasoning.",
          "url": "https://www.ulam.ai/"
        },
        {
          "kind": "rl-environments",
          "quote": "Train and evaluate agents in stateful, verifier-backed research environments where progress, revision, and failure become learning signals.",
          "url": "https://www.ulam.ai/"
        },
        {
          "kind": "datasets",
          "quote": "Curated open-problem corpora, structured metadata, and research workflows designed to probe reasoning beyond solved benchmarks.",
          "url": "https://www.ulam.ai/"
        },
        {
          "kind": "human-data",
          "quote": "Iterative traces where models explore, revise, and refine ideas with expert intervention at the points that matter.",
          "url": "https://www.ulam.ai/"
        }
      ],
      "areas": [
        "coding",
        "science",
        "math"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "These are synthetic or task-derived environments.",
          "url": "https://www.ulam.ai/data"
        },
        {
          "kind": "experts",
          "quote": "Iterative traces where models explore, revise, and refine ideas with expert intervention at the points that matter.",
          "url": "https://www.ulam.ai/"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "vals-ai",
      "name": "Vals AI",
      "website": "https://www.vals.ai",
      "what": "Benchmarks AI models and agents on real-world tasks in finance, coding, law and other areas, and publishes the results.",
      "says": {
        "quote": "Vals AI independently tests AI models on real-world work.",
        "url": "https://www.vals.ai"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "We run all of our own evaluations and create many of our benchmarks",
          "url": "https://www.vals.ai"
        },
        {
          "kind": "datasets",
          "quote": "A larger, privately held dataset, which we license to companies for their own internal validation.",
          "url": "https://www.vals.ai/methodology"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "finance",
        "legal",
        "medical",
        "math",
        "security",
        "robotics"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Developed in collaboration with domain experts, our datasets are carefully curated to be of extremely high quality",
          "url": "https://www.vals.ai/methodology"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "veris-ai",
      "name": "Veris AI",
      "website": "https://veris.ai",
      "what": "Sells simulation infrastructure that lets enterprises test, train and benchmark AI agents against simulated copies of their systems.",
      "says": {
        "quote": "A simulated copy of your systems, data, APIs, and users, where AI gets tested, trained, and built against before it ever touches production.",
        "url": "https://veris.ai"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "Build a benchmark around your own policies, workflows, and systems, then grade every agent you buy or build against it, end to end, in a twin of production",
          "url": "https://veris.ai"
        },
        {
          "kind": "rl-environments",
          "quote": "We train AI agents in high-fidelity simulations that accurately mirror their eventual production environments.",
          "url": "https://veris.ai/blog/introducing-veris-ai-a-new-way-to-train-enterprise-ai-agents-through-simulated-experience"
        }
      ],
      "areas": [
        "coding",
        "enterprise",
        "finance",
        "security",
        "speech"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "Scenarios, tool calls, and user interactions are all simulated, so no real customer data is needed.",
          "url": "https://veris.ai"
        },
        {
          "kind": "replica",
          "quote": "stateful twins of the systems the agent touches, simulated users pursuing goals, data that is consistent across all of it.",
          "url": "https://veris.ai/veris-for-benchmarking"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "verita-ai",
      "name": "Verita AI",
      "website": "https://verita-ai.com",
      "what": "Places subject-matter experts in remote, asynchronous projects for frontier AI companies.",
      "says": {
        "quote": "Remote, async engagements for experts with frontier AI companies, paid every Friday.",
        "url": "https://www.verita-ai.com/"
      },
      "sells": [
        {
          "kind": "human-data",
          "quote": "Remote, async engagements for experts with frontier AI companies, paid every Friday.",
          "url": "https://www.verita-ai.com/"
        }
      ],
      "areas": [],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Our platform solves this by collecting open-ended information about every expert in our community and building on each profile as they work with us.",
          "url": "https://verita-ai.com"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "vetto-ai",
      "name": "Vetto AI",
      "website": "https://vetto.ai",
      "what": "Designs RL environments, evaluations, expert data and human feedback for frontier AI labs.",
      "says": {
        "quote": "Vetto designs learning experiences, feedback systems, and evaluations that directly improve real-world model capability.",
        "url": "https://vetto.ai"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "Interactive environments where agents take on open-ended tasks and make real decisions, with verifiers built to reward the behavior that matters.",
          "url": "https://vetto.ai/companies/products.html"
        },
        {
          "kind": "evals",
          "quote": "Original tasks graded by a private verifier, comparing frontier models against failure modes, so labs find the next gap to close.",
          "url": "https://vetto.ai/companies/products.html"
        },
        {
          "kind": "human-data",
          "quote": "Data grounded in real expertise, from practitioners at the frontier of the sciences, medicine, law, finance, engineering, and the humanities.",
          "url": "https://vetto.ai/companies/products.html"
        },
        {
          "kind": "datasets",
          "quote": "Pre-built datasets spanning RL environments, coding, and core capabilities.",
          "url": "https://vetto.ai/companies/products.html"
        },
        {
          "kind": "annotation",
          "quote": "datasets that customers provide for annotation, evaluation, red-teaming, or related services",
          "url": "https://vetto.ai/companies/privacy.html"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "finance",
        "legal",
        "medical",
        "science",
        "vision",
        "speech",
        "multilingual"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Data across many languages, built by native experts who encode each one's grammar, idiom, and worldview, not translations.",
          "url": "https://vetto.ai/companies/products.html"
        },
        {
          "kind": "customer",
          "quote": "datasets that customers provide for annotation, evaluation, red-teaming, or related services",
          "url": "https://vetto.ai/companies/privacy.html"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "Menlo Park, USA",
        "quote": "1911 Valparaiso Ave, Menlo Park, CA 94025, USA",
        "url": "https://vetto.ai/companies/privacy.html"
      },
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "vibrant-labs",
      "name": "Vibrant Labs",
      "website": "https://vibrantlabs.com",
      "what": "Builds autoscaled reinforcement learning environments, tasks and evaluations for training and testing frontier AI agents.",
      "says": {
        "quote": "Post-training data that scales without armies of human annotators",
        "url": "https://vibrantlabs.com/"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "a suite of executable enterprise gyms for training and evaluating agents on realistic, policy-governed operational work",
          "url": "https://vibrantlabs.com/research/enterprise-worlds-itsmbench"
        },
        {
          "kind": "datasets",
          "quote": "Post-training data that scales without armies of human annotators",
          "url": "https://vibrantlabs.com/"
        },
        {
          "kind": "evals",
          "quote": "A benchmark testing how well coding agents visually clone real web apps through iterative pixel-diff feedback",
          "url": "https://vibrantlabs.com/"
        }
      ],
      "areas": [
        "coding",
        "computer-use",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "We use coding agents as world-builders that construct environments, tasks, and their verifiers",
          "url": "https://vibrantlabs.com/"
        },
        {
          "kind": "replica",
          "quote": "we built realistic, high-fidelity simulated replicas of email and calendar web applications",
          "url": "https://vibrantlabs.com/research/pa-bench"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Vibrant Labs has 6 employees based in San Francisco",
        "url": "https://www.ycombinator.com/companies/vibrant-labs"
      },
      "founded": {
        "value": 2023,
        "quote": "Founded in 2023 by Shahul ES and Jithin James",
        "url": "https://www.ycombinator.com/companies/vibrant-labs"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "visionlab-ai",
      "name": "VisionLab AI",
      "website": "https://thevisionlab.ai",
      "what": "Captures, structures, and annotates real-world industrial video and process data for frontier AI labs and robotics companies.",
      "says": {
        "quote": "The company provides real-world industrial data to frontier AI labs and robotics companies through a global network of more than 2,000 factories.",
        "url": "https://thevisionlab.ai/articles/seed-2026"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Its datasets include both egocentric footage, showing how trained workers perform tasks, and exocentric footage, capturing broader factory context, workflows, tools, and operating conditions.",
          "url": "https://thevisionlab.ai/articles/seed-2026"
        },
        {
          "kind": "annotation",
          "quote": "Through these partnerships, the company captures, structures, and annotates real-world industrial video from active production environments.",
          "url": "https://thevisionlab.ai/articles/seed-2026"
        },
        {
          "kind": "human-data",
          "quote": "Robotics teams need access to real industrial environments, real workflows, and real human task execution at scale.",
          "url": "https://thevisionlab.ai/articles/seed-2026"
        }
      ],
      "areas": [
        "robotics",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "captured",
          "quote": "Through these partnerships, the company captures, structures, and annotates real-world industrial video from active production environments.",
          "url": "https://thevisionlab.ai/articles/seed-2026"
        },
        {
          "kind": "experts",
          "quote": "atomic action annotations with full expert validation",
          "url": "https://thevisionlab.ai"
        }
      ],
      "buysData": null,
      "hq": {
        "value": "San Francisco",
        "quote": "Vision Lab is headquartered in San Francisco.",
        "url": "https://thevisionlab.ai/articles/seed-2026"
      },
      "founded": {
        "value": 2025,
        "quote": "Founded in 2025 by Tanachart (James) Kujareevanich, Zhichu Ren, and Will Wenbo Zhang, Vision Lab has 11 employees based in San Francisco.",
        "url": "https://www.ycombinator.com/companies/vision-lab"
      },
      "readOn": "2026-10-15"
    },
    {
      "id": "vmax",
      "name": "Vmax",
      "website": "https://vmax.ai",
      "what": "Builds reinforcement learning environments for training and evaluating language-model shell agents.",
      "says": {
        "quote": "We introduce unix-ctf, a procedural generator of capture-the-flag environments for training and evaluating Unix competence in language-model shell agents.",
        "url": "https://vmax.ai/campaign"
      },
      "sells": [
        {
          "kind": "rl-environments",
          "quote": "We introduce unix-ctf, a procedural generator of capture-the-flag environments for training and evaluating Unix competence in language-model shell agents.",
          "url": "https://vmax.ai/campaign"
        },
        {
          "kind": "evals",
          "quote": "We introduce unix-ctf, a procedural generator of capture-the-flag environments for training and evaluating Unix competence in language-model shell agents.",
          "url": "https://vmax.ai/campaign"
        }
      ],
      "areas": [
        "coding",
        "math"
      ],
      "madeFrom": [
        {
          "kind": "synthetic",
          "quote": "It procedurally generates capture-the-flag tasks inside fresh Linux containers.",
          "url": "https://vmax.ai/team/unix-ctf-procedural-environments-for-unix-competence-reinforcement-learning"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "wirestock",
      "name": "Wirestock",
      "website": "https://wirestock.io",
      "what": "Sells licensed multimodal image, video and creative datasets, including custom datasets, to AI companies.",
      "says": {
        "quote": "The platform connects creators with the world’s leading AI companies, enabling the continuous creation, curation, and licensing of creative multimodal data that trains and improves advanced models.",
        "url": "https://wirestock.io/about-us"
      },
      "sells": [
        {
          "kind": "datasets",
          "quote": "Datasets built for pretraining, SFT, RL, and RLHF workflows.",
          "url": "https://wirestock.io"
        },
        {
          "kind": "human-data",
          "quote": "Structured to your specifications, with rubrics, verifier tasks, and expert QA and review.",
          "url": "https://wirestock.io"
        }
      ],
      "areas": [
        "creative",
        "vision"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "Premium multimodal data from a global network of creative professionals.",
          "url": "https://wirestock.io"
        },
        {
          "kind": "licensed",
          "quote": "Upload your work once and earn commission whenever it is licensed.",
          "url": "https://wirestock.io"
        }
      ],
      "buysData": {
        "quote": "Creators are compensated for every creative contribution as it is licensed to power industry-leading AI models.",
        "url": "https://wirestock.io/about-us"
      },
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    },
    {
      "id": "xdof",
      "name": "XDOF",
      "website": "https://www.xdof.ai/",
      "what": "Runs real-world robot policy evaluations and delivers robot training data for robotics labs and AI companies.",
      "says": {
        "quote": "We run rigorous real-world policy evaluations for teams that need to build faster.",
        "url": "https://www.xdof.ai/products/evaluations"
      },
      "sells": [
        {
          "kind": "evals",
          "quote": "We run rigorous real-world policy evaluations for teams that need to build faster.",
          "url": "https://www.xdof.ai/products/evaluations"
        },
        {
          "kind": "datasets",
          "quote": "delivering the highest-quality teleoperation, on-embodiment, and egocentric data",
          "url": "https://www.xdof.ai/careers/forward-deployed-engineer-fde-cd508d93-5ae9-486a-a633-48402ef1fe53"
        },
        {
          "kind": "annotation",
          "quote": "Ensure data annotation projects meet client requirements on accuracy and SLA, while satisfying internal cost targets.",
          "url": "https://www.xdof.ai/careers/strategic-project-lead-574477cf-6fa1-4644-8927-67cfe47b015f"
        }
      ],
      "areas": [
        "robotics",
        "enterprise"
      ],
      "madeFrom": [
        {
          "kind": "experts",
          "quote": "We run your policies across our global robot fleet, with expert operators trained in the exact tasks being run.",
          "url": "https://www.xdof.ai/products/evaluations"
        },
        {
          "kind": "captured",
          "quote": "delivering the highest-quality teleoperation, on-embodiment, and egocentric data",
          "url": "https://www.xdof.ai/careers/forward-deployed-engineer-fde-cd508d93-5ae9-486a-a633-48402ef1fe53"
        }
      ],
      "buysData": null,
      "hq": null,
      "founded": null,
      "readOn": "2026-10-15"
    }
  ]
}