Find companies working with business data

6thSense makes a wearable tactile sensor and a first-person stereo camera, and sells the synchronized touch, video and motion data it records from people doing hand tasks to teams training dexterous robots.

Abundant builds reinforcement-learning environments, evaluation benchmarks, and training datasets for AI labs, AI startups, and large enterprises developing and improving AI agents.

AfterQuery sells expert-created and synthetic training datasets, reinforcement-learning environments, and custom evaluations to AI labs and enterprises developing foundation models and agents.

AI Astrolabe supplies human-written and annotated training datasets, model evaluations and red-teaming services, with a focus on Arabic dialects and other low-resource languages. Its catalog includes text, speech and audio, images, and safety preference data.

AIChamp provides robotics and AI teams with proprietary simulated environments and physics-verified evaluations for robot policies, running large-scale trials and returning scores, videos, trajectories, and failure analysis before real-world deployment.

Alice, formerly ActiveFence, sells AI safety and security training datasets, agentic RL environments, and expert-led red teaming and evaluations to foundation model labs; it also sells WonderSuite guardrail software to enterprises and content moderation to online platforms.

Align AI sells expert-grounded training data, simulated RL environments, continuous evaluations, and in-runtime verification tools to AI labs and enterprises building agents for regulated, high-stakes work.

Andon Labs sells custom evaluations and benchmarks to AI labs and developers, testing AI agents in simulations and real-world deployments across business operations, spatial reasoning, computer use, and robotics.

Andromede provides AI labs and agent developers with reinforcement-learning environments, tasks, and verifiers built from real-world data for training and evaluating long-horizon agents.

Anthromind supplies AI developers, research teams, and businesses with expert-curated training and preference data, custom LLM evaluations, and fine-tuning support through tailored projects and managed data-curation services.

Apex Data Sciences, the AI division Apex CoVantage launched in 2024, sells expert RLHF and SFT data, expert-built benchmarks and image, text, video and lidar annotation to AI teams. Its parent also sells publishing, accessibility and healthcare billing services.

Appen supplies AI labs and product teams with human-produced training data, annotation, expert feedback, RL environments, and model evaluations for machine-learning and agentic AI systems.

Applied Intuition sells simulation, data and AI-development software, vehicle operating systems, and autonomy tools to automakers, defense organizations, and other teams building intelligent machines.

Aptura provides expert-built training and evaluation data, benchmarks, and reinforcement-learning environments to AI labs and developers building systems for finance, healthcare, and legal work.

Archal sells resettable API sandboxes that mimic third-party software services, giving AI developers and software teams repeatable environments to test, evaluate, and train agents without using live accounts.

Arena sells AI model evaluations built on real-world human feedback to labs, enterprises and developers, with analytics, feedback samples and SLA-backed delivery. Its privacy policy says it may share users' prompts, votes and other content with AI model developers to evaluate, improve and develop their products; its FAQ says such shared data is de-identified.

Arga provides AI developers and engineering teams with stateful replicas of business software, letting them train and test agents on realistic workflows and validate changes without modifying live production systems.

Builds production-style simulated workspaces and verifiable benchmarks for AI agents, helping AI developers train and test long-horizon tasks in cybersecurity, incident response, software engineering, and tool use.

Asimov collects first-person video and sensor data of people doing everyday tasks in real homes and workplaces, and delivers it as annotated datasets to robotics teams training humanoid robots.

Aviro sells reinforcement-learning environments, post-training datasets, and evaluations for AI labs and teams building long-running agents, with tasks spanning coding, computer use, and enterprise knowledge work.

Builds simulated, workflow-based environments and benchmarks, plus a runtime and managed post-training service, for AI labs and developers training and evaluating agents.

Besimple AI supplies AI labs and voice-AI developers with licensed conversational audio, expert annotation, and benchmarks for training and evaluating speech models.

Bespoke Labs sells realistic, company-scale reinforcement-learning environments and agent-training, optimization, and evaluation infrastructure to frontier AI labs and enterprises.

BioStack sells clinical and preclinical data, healthcare AI training environments, and evaluation tools to AI labs and life-sciences teams developing medical and drug-discovery models.

Bugcrowd provides cybersecurity training environments and an exploitation benchmark to AI developers, alongside crowdsourced vulnerability testing services for organizations.

Build AI records first-person video of workers doing manual jobs in real factories and businesses, and supplies it to physical-AI and robotics labs as an off-the-shelf training dataset or as data tailored to a lab's request.

Boston company that labels medical and other data through scored contests among human annotators, specialists and AI models, sells model evaluation services, and licenses finished labeled datasets, such as chest X-ray and pathology sets, to AI developers.

Centific sells multimodal AI training data, human annotation, model evaluations, and simulated enterprise RL environments to frontier AI labs and enterprise AI teams.

Chakra Labs builds expert-authored tasks, graders and computer-use environments for training and evaluating AI models. Its off-the-shelf and bespoke task sets cover finance, legal work, responsible AI and computer use.

Cint collects commissioned human responses in text, audio and video for AI-model training and validation through its respondent network and AI-moderated interview workflows. This profile covers newly created survey and conversational data; its separate Verified Audiences consumer-profile licensing product is outside this listing’s scope.

Claru provides custom and licensed video, robotics demonstrations, and human annotations to AI labs building robotics, embodied AI, and world models.

CloudFactory supplies a managed human workforce and AI platform to businesses and AI developers, collecting and annotating data, preparing model-training and fine-tuning datasets, and providing human feedback, evaluation, and operational support for AI systems.

Abu Dhabi company that sources, labels and evaluates training data for AI labs, hyperscalers and robotics teams, including human-ranked model responses, expert-written evaluation sets, Arabic speech data and first-person video of real work filmed by paid contributors through its KLKT app, and also sells its own Arabic speech models and voice agents.

Collinear sells stateful simulated environments, agent evaluations, and synthetic training data to frontier AI labs and enterprise AI teams building and improving AI agents.

Consequence Labs creates custom human-generated training and evaluation datasets, including workflow demonstrations, speech data, expert judgments and annotations. It recruits contributors, manages collection and review, and delivers structured data with provenance and quality documentation.

Contra Labs provides AI teams with professional creatives’ judgments, model evaluations, benchmark datasets, human-feedback training data, and recorded design workflows to improve and measure generative AI.

Cortex AI provides AI labs with real-world workplace video and robot trajectories for training embodied AI, plus human-supervised robot rollouts and evaluations that capture recovery from failures.

Cua sells cloud and self-hosted desktop sandboxes, computer-use runtimes, RL environments and benchmarks, and trajectory data to AI developers building, training, and evaluating agents that operate computers.

CueBench builds reinforcement-learning environments and task benchmarks for AI labs and developers to train and assess frontier models on difficult scientific-reasoning and performance-engineering problems.

DARE Labs offers Cerebro, a data refinery that turns operational data into supervised-training datasets, RL data and environments graded on real outcomes.

Datacurve sells expert-built coding data, reinforcement-learning environments, benchmarks, and evaluations to foundation-model labs and AI developer-tool companies.

Dataloop sells an AI data-development platform and on-demand annotation services to enterprises and AI teams, helping them organize, label, quality-check, and route multimodal data into AI training and production workflows.

DataOcean AI sells off-the-shelf datasets and custom data collection, labeling, and model evaluation services to AI developers and enterprises, supporting LLM, speech, vision, multimodal, and autonomous-driving systems.

Datoric licenses robotics and physical-AI training data (egocentric video, manipulation and teleoperation recordings), runs custom collections to a buyer's specification, and also offers computer-use, voice and gameplay datasets.

David AI designs, collects, annotates, and licenses human-recorded audio datasets, and evaluates speech models for AI labs and developers building speech recognition, translation, synthesis, and conversational systems.

Deccan AI sells AI labs and developers expert-created post-training datasets, human feedback, reinforcement-learning environments and evaluations; it also offers enterprises monitoring and AI-enabled operations tools.

Deeptune builds simulated workplace-software environments, bundled with tasks, datasets and infrastructure, where frontier AI labs train agents to carry out multi-step digital work. Mercor acquired Deeptune in July 2026.

Defined.ai sells licensed, multilingual AI training datasets and provides custom data collection, annotation, evaluation and model fine-tuning services to organizations developing AI.

Diffuse Labs builds reinforcement-learning environments for frontier AI labs, generating scalable tasks and verifiable rewards so developers can train agents and measure gains across coding, chip design, Kubernetes operations, and enterprise software.

Dissei sells finance-focused RL environments, evaluation benchmarks, and real-transaction training data and analyst trajectories to AI labs and model developers, as well as financial institutions building or assessing models.

d_model builds RL environments where agents learn to do interpretability and alignment research, and runs similar agents in its own research.

e2f, which renamed itself TrustScale in June 2026, sells custom AI training datasets, expert annotation, model evaluation and benchmarking, and reinforcement learning feedback through its ArgusRL platform, covering 200 languages and text, image, audio and video data.

EdotEnv provides financial-market-based reinforcement-learning environments and research benchmarks that AI labs and research teams use to train and assess agents on quantitative-research tasks.

Eigent sells an open-source multi-agent desktop workforce for automating enterprise workflows, and builds custom RL environments, benchmarks, and training/evaluation loops for enterprise teams and AI developers.

Emulated offers AI teams post-training datasets, production-like software-engineering environments, and benchmarks to train and evaluate agents on coding, infrastructure operations, machine-learning engineering, and research tasks.

Enact works with robotics model developers to identify failures on physical robots, create targeted recovery demonstrations and training data, then fine-tune and evaluate models for real-world tasks.

Encord sells a multimodal data platform and managed collection, annotation, and evaluation services to AI teams preparing and improving training data for physical AI, computer vision, and generative models.

Supplies large-scale pretraining and post-training datasets, including code and web, document, and video data, plus verifiable STEM RL environments and custom data sourcing for frontier AI labs.

Exponential sells AI developers multi-agent security environments for RL post-training, security-focused training data, and adaptive red-team evaluations that test AI models and agent systems against realistic attacks.

Fleet sells simulated training environments and evaluation tools where AI agents practice realistic tasks to frontier AI labs, hyperscalers, and enterprises.

Flevy, a business-template marketplace, runs Flevy Business Data Services, which catalogs documents held by firms and professionals and pitches that inventory to AI labs for licensing; it also connects labs with contributors for consulting and evaluation work.

FoundryAI creates expert-reviewed coding and other technical-workflow data, RL-ready task suites, private model evaluations, and regression tests for AI development teams.

G2i supplies coding post-training data to AI labs (RL environments, evals, RLHF preference data, SFT datasets, scrubbed seed data) through managed projects or contract engineers placed on lab teams. It also runs an engineer hiring platform and tech conferences.

General Reasoning offers AI labs and developers API-hosted reinforcement-learning environments, plus infrastructure to deploy environments and train or evaluate agents at scale.

Through GLG Frontier, a practice it formally launched in October 2026 after working with frontier labs since 2023, GLG recruits credentialed professionals in medicine, science, law, finance and engineering to design model evaluations, write tasks and grading rubrics, review model outputs and red-team models for AI labs. Its main business is an expert network that connects investors, consulting firms, companies and law firms with industry experts, mostly through calls.

Good Start Labs builds game-based reinforcement-learning environments, turns gameplay into human and synthetic training data, and publishes evaluation benchmarks for AI labs. It also helps game publishers capture gameplay data and put AI agents into their games.

Gradient Health sells de-identified medical imaging and multimodal clinical datasets through Atlas to medical AI developers, helping them train, test, validate, and prepare regulatory evidence for healthcare models.

Gray Swan sells AI-model and agent security to frontier labs and enterprises: adversarial testing before release, plus runtime filtering and monitoring to block attacks, unsafe outputs, and policy violations.

Pune-based company that offers benchmark licensing and custom model evaluations to AI labs and enterprises, and publishes BFSI-Bench, an Indian financial-regulation benchmark. It also creates, labels and reviews multimodal training data and screens egocentric video for embodied AI.

Halluminate sells finance-focused computer-use and tool-use RL environments, simulated business applications, training data, and agent benchmarks to frontier AI labs and enterprise AI teams.

Handshake supplies frontier AI developers with expert-created training and evaluation data, reinforcement-learning environments, and licensed, de-identified operational records from real companies.

Hillclimb creates custom mathematics-focused reinforcement-learning environments and supervised fine-tuning data from expert problem-solving, selling them to frontier AI labs to train models and research agents.

Hiloop sells automated research infrastructure to AI/ML teams, running experiments across their models, data, training methods, prompts, tools, and systems against customer-defined evaluations, then returning verified improvements.

HUD sells an SDK and cloud platform for building, running and grading reinforcement-learning environments, turning agent runs into evaluation results and training data. Through its DataVendor marketplace, organizations license private data such as codebases, tasksets and environments to invited AI buyers.

Human Archive collects and labels real-world human video and synchronized sensor recordings, then licenses multimodal datasets to AI labs and robotics developers training physical-AI systems.

Humyn Labs collects and annotates real-world multimodal human data—especially first-person video, sensor and motion signals, and speech—and supplies training datasets and evaluation benchmarks to AI labs, robotics developers, and other AI teams.

Huzzle Labs sells reinforcement-learning environments, expert-produced training data, evaluations, and custom-trained models to AI labs and enterprises seeking agents that can handle real software and business workflows.

HyperPrior Research is a small London research firm that builds environments, graders and agent test suites for teams training and evaluating AI agents, sold as fixed-scope or retainer engagements alongside technical diligence, strategy and machine-learning consulting.

Idler sells reinforcement-learning environments, benchmarks and evaluation datasets based on production work to AI labs and other teams training and evaluating models.

iMerit sells expert-led data annotation, training-data creation, human feedback, model fine-tuning and evaluation services, alongside Ango Hub workflow software, to frontier AI labs, AI developers and enterprises building multimodal and domain-specific models.

Incalmo builds reinforcement-learning cyber environments and security evaluations for AI labs and security teams to train and assess cybersecurity-capable models.

Innodata provides AI developers and enterprises with training data, human feedback, annotation, reinforcement-learning environments, model evaluations, and red-teaming services for building and improving AI models and agents.

Intelligence runs DesignArena, where people vote between AI-generated websites, games, images, video and audio, and sells model evaluations to AI labs. It says it shares users' prompts, outputs and votes with AI model developers, who may train on them, after taking steps to remove identifiers like usernames; its terms let it charge for that use.

Invisible sells expert-created training data, model fine-tuning, evaluations, and custom reinforcement-learning environments to AI labs and developers; it also provides AI software and implementation support for enterprises.

A marketplace that curates and licenses human-contributed and rights-holder datasets—including video, images, audio, and text—for AI labs, research institutions, and enterprises training models.

Labelbox sells simulated reinforcement-learning environments, expert-generated preference and training data, and evaluation services to frontier AI labs; it also offers enterprise teams software for running agents on real company workflows.

LatchBio sells cloud data and compute software, bioinformatics workflows, and AI-based analysis to biotech R&D teams, laboratories, and scientific kit or instrument providers. It also publishes evaluations and offers assessment services for biology-focused AI agents.

Lightwheel sells physics-accurate simulation assets and environments, synthetic and human demonstration data, and evaluation tools to robotics and Physical AI teams training and validating robot models.

LightWrk sells expert-produced reasoning annotations, preference and training signals, and evaluation workflows to robotics, physical-AI, and world-model teams.

Luel licenses ready-made multimodal datasets and organizes custom collection campaigns for AI labs and developers, using paid contributors to supply rights-cleared audio, video, speech, sensor, and computer-use data.

LXT supplies AI developers and enterprises with custom and off-the-shelf training data, human annotation, evaluation, and RLHF services across text, speech, images, and video.

Macrodata processes robot and first-person video for robotics and embodied-AI teams, returning timestamped action labels, 3D hand tracks, task-progress signals, and quality checks that help prepare data for model training.

Connects companies’ existing agent-data systems, collects approved interaction traces, corrections, and completed workflows, then anonymizes and packages them as licensed training data or environments for AI labs and researchers; data suppliers earn a revenue share.

Maingen builds industrial-operation simulations, RL environments and benchmarks for frontier AI labs, helping them train and assess agents on tasks such as running solar portfolios and factories.

Markov sells computer-use datasets and task environments—including expert CAD, productivity, coding, and gaming workflows—to AI labs and developers training or evaluating desktop agents.

Matrices sells realistic simulated computer-use environments to AI labs, where multimodal agents practice browser and workplace tasks through reinforcement learning without real-world side effects.

Maxinsights collects and annotates real-world egocentric video and multimodal human-motion data for robotics and physical-AI developers, and markets simulation and model-evaluation capabilities for robot and world models.

Mechanize builds simulated software-engineering tasks and benchmarks for frontier AI labs, helping them train and evaluate coding agents using graded performance.

Mecka collects and structures real-world human activity into robotics training datasets, and provides data tools and evaluation and deployment support for robotics labs, frontier AI teams, and enterprises building embodied systems.

Mercor sells expert-created training data, reinforcement-learning environments, and evaluations and benchmarks to frontier AI labs and other AI developers. It also offers enterprise data licensing and evaluation services.

Metaphi sells custom and reusable simulated work environments, training datasets, and agent evaluations to frontier AI labs and enterprise AI teams, helping them train and assess agents on realistic software, finance, and content-generation tasks.

micro1 provides expert human data and post-training services, real-world RL environments, robotics data, and AI-agent evaluations for frontier AI labs, enterprises, and robotics teams.

microagi sells datasets of recorded human manual work to AI labs and runs Atlas, a service that records a factory's own tasks, trains robot models on that data and installs the robots on the line.

Midcentury sells real-world human-action and gameplay datasets, along with cloud simulation tools, to robotics developers, AI labs, and enterprises training and evaluating physical AI systems.

Mind Lab provides pharmaceutical, financial, public-sector and gaming companies with custom benchmarks, task environments, data development and model post-training, alongside its own training platform.

Miro Advisory helps established businesses license their operating history (support tickets, reconciliations, dispatches, approvals and other workflow records) and software companies license private codebases with their commit, issue and review history, to AI developers for training and evaluation. It acts as an advisor and intermediary: it handles qualification, rights review, buyer discussions and the licensing process.

Miso Motion runs custom video collection for physical AI and robotics teams, recording consented professionals on paid jobs with one to six synchronized cameras and delivering redacted footage with per-clip records and optional task labels.

Mode buys copies of company-owned operating records, removes sensitive material or transforms personal identifiers, and resells the prepared data for AI research and training. This profile covers its company-data offering; Mode separately operates a consumer-data business supplying AI buyers.

Monte sells post-training and continual-learning infrastructure to AI labs and enterprise AI teams, helping them evaluate models against work tasks, train specialized agents, and improve them using feedback from real workflows.

Moonlake provides world-model and simulation tools that turn real-world inputs or text into simulation-ready 3D assets and environments for robotics, autonomy, and AI research teams.

Navaday supplies expert-created training data and model evaluations across professional and technical fields. It manages specialists who write examples, correct existing data, compare model answers and produce held-out test sets. It also prepares and helps license business records for AI development.

NewtonX recruits professionals in finance, tech, healthcare and law, checks their identity and expertise, and uses them to run model evaluations, produce post-training datasets and audit annotation quality for AI developers. Its main business is B2B market research.

Normal publishes CAD benchmarks, datasets, and technical reports for AI developers working on hardware design; CAD Arena evaluates how agents reconstruct real mechanical parts from engineering drawings as editable CAD models.

Noumad builds formally verified reinforcement-learning environments for post-training AI agents on unsolved problems in long-horizon machine learning, computational science, robotics and cybersecurity.

Nyne AI licenses workflow data from companies, de-identifies it and prepares training and evaluation datasets for AI labs, and connects labs with domain specialists. It also sells people and company data APIs built on its identity graph.

Olam Labs builds simulated multi-agent environments, behavioral evaluations and game-derived datasets for AI labs and researchers training or evaluating models’ social, agentic and safety capabilities.

Ooak Data licenses and anonymizes real company workflows, then turns them into RL training and evaluation environments for frontier AI labs and teams building enterprise agents.

OpenTrain connects AI labs and developers with vetted human experts for RLHF, model evaluation, red teaming, and data annotation. Customers hire directly or use a managed team working in their existing tools.

Creates data intended to help AI systems operate legacy software, targeting AI labs. Its public website invites interested users to join a waitlist.

Paper Instruments post-trains language models and builds evaluation benchmarks, post-training infrastructure, and agent software for AI developers working on knowledge-intensive tasks.

Pareto supplies AI labs with specialist-created training data, evaluations, and RL environments that turn expert judgment into signals for assessing and improving frontier models.

Parsewave creates custom, expert-written datasets, solution traces, and evaluations for frontier AI labs and developers to train and test coding and other professional-domain models.

Patronus AI builds training data and simulated environments to train and improve AI models and agents. Its research studies model failures in alignment, coding, knowledge work, science and computer use, and it publishes benchmarks.

Perle supplies AI labs and developers with specialist-generated multimodal training data, annotation, RLHF and model evaluations through expert-led programs and managed workflows.

Petrarch sources operational records from companies under signed agreements, de-identifies them with its in-house Lindo software inside the supplier's environment, and licenses or sells the prepared datasets to AI labs and applied AI startups as training data.

Phinity builds chip-design reinforcement-learning environments and training infrastructure for AI labs, while developing agents to design, verify, and eventually tape out custom chips.

Pixelstwo sells curated tactile and dexterous manipulation datasets, captured with proprietary tactile hardware, to robotics teams training robots for contact-rich tasks.

Plato builds simulated software and computer environments from real-world data so AI agent developers can train, test, and score browser and computer-use agents.

Poindexter Labs supplies expert data annotation and evaluation services to frontier AI teams, alongside syncronus, a collaborative platform for expert-led model training.

Builds simulated environments, tasks, and verifiers for AI labs to train and evaluate long-horizon agents; its Horizon-SWE benchmark tests end-to-end software engineering workflows.

pre.dev sells runnable, graded coding environments and training and evaluation data built from licensed private software repositories to AI labs and developers training coding agents.

Preference Model builds reinforcement-learning environments for frontier AI labs, helping train models to handle machine-learning research and engineering tasks.

Prime Intellect sells AI labs and developers an open stack for reinforcement-learning environments, model evaluations, hosted training, inference, sandboxes, and GPU compute to build and improve agentic models.

Prolific gives AI labs and developers access to verified participants and subject-matter experts to collect human feedback, training data, and model evaluations through a platform, API, or managed research service.

Protege licenses data from hundreds of partner organizations, curates it and de-identifies health records, and licenses the resulting datasets to AI developers for pre-training, post-training and evaluation across healthcare, video, audio and speech, and spatial and physical data. It also runs independent, double-blind evaluations of healthcare AI models on de-identified patient records.

Provision supplies real-world multimodal training data, physical-world evaluations, and deployment support for teams developing physical AI and robotics.

Proximal builds coding tasks, training data and RL environments that AI labs and enterprises use to train software-engineering agents.

Quesma publishes coding-agent benchmarks and builds RL training environments for frontier labs; it also markets tools that help development teams understand agent costs, workflows and outcomes.

ReadySolve licenses MartechBench, an enterprise marketing-workflow environment for reinforcement learning and evaluation. Agents operate connected business systems with deterministic task verification; buyers receive runtime materials and can commission model benchmarks.

ReasonCore provides expert-curated AI training datasets, reinforcement-learning environments, and adaptive evaluations for frontier AI labs, emerging AI labs, and enterprises developing or deploying advanced models.

Refresh sells simulated software environments and verifiable evaluations for training and testing AI agents, alongside real-workflow coding datasets, to frontier AI labs and enterprises.

Replay pays companies for their work history (code, internal docs, support tickets, CRM and financial records), strips personal identifiers and licenses the cleaned datasets to frontier AI and data labs. The same company runs Sunset, a startup wind-down service.

Revelo sells AI labs code-focused training data, evaluations, benchmarks, and RL environments, with expert-led data creation and managed workflows to help improve coding models and agents.

Rise Data Labs sells human- and synthetic-data production, annotation, and model-evaluation services to AI labs and enterprise model teams building and improving generative AI.

RL Labs builds long-horizon software-engineering environments where frontier AI labs can train models on complex, multi-step coding tasks.

Runloop provides isolated development sandboxes, orchestration, and benchmarking tools so AI developers and labs can build, run, score, and refine coding agents, including reinforcement-learning and fine-tuning workflows.

Sama sells human-in-the-loop data annotation, validation, evaluation, and training-data services to enterprise AI teams building computer-vision, multimodal, and generative AI models.

San Francisco Data prepares and licenses training data derived from companies’ production AI traces, starting with coding agents. Its research work turns recorded interactions into reproducible tasks and training environments for frontier models.

Sapien supplies human-created and expert-reviewed training data to AI developers, including data annotation, collection, curated datasets, and quality checks for AI outputs.

Scale sells AI labs and businesses custom training data, human feedback, annotation, RL environments, and model evaluations, alongside software to build, test, and deploy AI agents and applications.

Sepal supplies frontier AI labs and enterprise model teams with expert-created training and evaluation data, benchmarks, and reinforcement-learning environments for developing and testing models on specialized tasks.

San Francisco company that builds finance and coding RL environments, with tasks created by private equity professionals and senior software engineers, and sells ready-made environments, custom tasks, model evaluations and human preference data to frontier AI labs.

Sieve supplies AI labs and model developers with curated or custom-collected video, audio, image, and interaction data, plus annotations, evaluation sets, and interactive environments.

SimpleClosure’s AssetHub connects startups with AI buyers interested in licensing or acquiring source code and workspace data.

Snorkel sells expert-built training datasets, agent environments, benchmarks and evaluations—through ready-to-use products and custom data-development services—to frontier AI labs and enterprise AI teams.

Sticker builds reinforcement learning environments and benchmarks for training and evaluating AI agents on procurement work, including backorders, purchasing policies and mismatched invoices, for teams post-training models. It previously built and ran procurement infrastructure for businesses.

SuperAnnotate sells AI data software and managed expert services to AI labs and developers, covering multimodal annotation, training data, model and agent evaluations, and reinforcement-learning environments.

Surge sells AI labs and developers expert-built training data, human feedback, evaluation benchmarks, and interactive reinforcement-learning environments to train and assess AI models and agents.

Symbal supplies AI labs with multimodal post-training datasets, private evaluations and benchmarks, and human-in-the-loop data; it also builds enterprise agents using customers’ first-party workflows.

An AI co-scientist workbench for research teams, automating literature review, hypothesis generation, experiments, analysis, and paper drafting; the company also develops scientific evaluations and long-horizon RL environments for AI labs.

Tacit builds expert-authored life-science evaluations, scientific agent environments, and structured datasets to help frontier AI labs train and assess models on long-horizon biology, chemistry, and drug-development tasks.

TaskUs provides outsourced AI data operations for AI developers and enterprises, including data collection and labeling, expert feedback and RLHF, model evaluations and red-teaming, and robotics data services.

Taste Labs sells preference and reasoning data, rubrics, and evaluation environments to frontier AI labs for subjective design quality, and provides agent and app developers a Brand API and MCP tools to extract, search, and verify brand systems.

Telegraph Lab sources and licenses business records for AI training. It reviews data rights, prepares datasets through de-identification and structuring, and arranges delivery to AI labs. Its sources include workplace communications, business-system records, company files and operating histories.

TELUS Digital supplies AI labs and developers with expert-built training data, annotations, agentic RL environments, red-team testing and evaluations, plus custom multimodal and physical-AI datasets and managed data-collection services.

TensorTest sells private evaluations, reinforcement-learning environments and expert-made training data to AI labs building creative agents, starting with video editing; its public TimelineBench benchmark scores agents on 56 editing jobs.

The General Data Company licenses anonymized records and workflow-based datasets from European businesses to AI labs for model training and evaluation, including held-out benchmarks and reinforcement-learning environments.

Theta sells custom simulated work environments and hands-on implementation to AI labs and enterprises, helping them train and evaluate specialized agents on realistic, multi-step business workflows.

thetalab builds on-prem agent-training infrastructure for AI developers. It turns customer production traces and difficult workflow cases into private RL environments, then uses the results to train custom models for specific business processes.

Toloka supplies AI labs and developers with human-expert and synthetic training data, agent environments, evaluations, and red-teaming to build and assess AI models and agents.

Trajectory Labs runs safety evaluations, expert red-teaming and RL environments that help AI labs find model vulnerabilities and measure safer behavior.

Traverse builds reinforcement-learning environments and training data for frontier AI labs, helping them train models on ambiguous professional work that depends on judgment, context, and long-horizon reasoning.

Troveo licenses and packages rights-cleared real-world data—including video, audio, text, business workflows, gaming and robotics—for AI labs and model developers, with custom sourcing and annotation.

Turing supplies AI labs with training data, evaluations, and RL environments for coding, enterprise knowledge work, and STEM. It also helps large enterprises build and deploy AI agents in important workflows.

Uber AI Solutions provides custom AI data collection and annotation, specialist human feedback and RLHF, red-teaming, and model evaluation for frontier labs, enterprises, robotics firms, and autonomous-vehicle developers.

Ulam sells reasoning datasets, expert-reviewed traces, benchmarks, and verifier-backed RL environments—primarily for mathematics and STEM—to AI labs and developers.

Vals AI sells evaluation software, expert-built benchmarks, and review tools to AI labs and enterprise engineering teams testing models and AI applications on practical, domain-specific tasks.

Veris sells AI developers a simulation and reinforcement-learning platform that recreates enterprise tools, systems, and users, helping teams train, test, and improve agents on realistic workflows before production.

Verita supplies vetted domain experts and manages project operations for AI labs, producing annotation, human-feedback and evaluation work, training data, and support for reinforcement-learning environments.

Vetto sells AI labs and developers expert post-training data, human evaluation, custom RL environments, and verifier-graded benchmarks to help train and assess models on practical tasks.

Vibrant Labs builds RL-ready simulated environments, training tasks, and benchmarks for AI labs and developers working on long-horizon agents, including agents that use browsers, computers, or enterprise tools.

Vision Lab sells real-world factory video and SOP-linked annotations to AI and robotics teams, giving them structured examples of human tasks and industrial processes for training physical-AI models.

Vmax markets Campaign, a self-play task-generation and RL platform that uses developers’ proprietary data and evaluations to create training environments and refine agents on new task examples.

Watt sources, prepares and licenses private operational datasets for AI training and evaluation, with a current emphasis on energy and industrial records. This profile covers that company-data offering; Watt also operates a behavioral-data Signal Graph, and its bulk catalog includes intent, employment and contact data. Consumer-profile datasets are outside this profile.

Wirestock supplies AI labs with licensed, creator-made multimodal data, offering catalog datasets and custom collections for training AI models.

XDOF collects robot teleoperation data and runs real-world policy evaluations for robotics labs and AI companies, and also supplies robotic systems and data tooling.

Xplore AI builds expert-designed reinforcement learning environments and evaluations for mechanical, electrical and aerospace engineering. Its RWE-bench tests agents using professional engineering tools and deterministic verifiers; separate environments are available for training.