training-data
(14)
Bright Data Is Building a Toll Road for the Machine-Readable Web
What began as proxy infrastructure now sells something AI agents urgently need: live public web data that arrives structured, usable and less likely to break at the first CAPTCHA.
6thSense
6thSense builds wearable tactile sensing rigs that record how human hands actually touch the world - pressure, grip, contact, force, depth, vision and 3D hand pose - out in real homes rather than labs. It packages that multimodal, model-ready data and sells it to teams training the next generation of dexterous robots, positioning itself as a 'nervous system for physical AI.' The company is part of Y Combinator's Summer 2026 batch and is based in San Francisco.
Mundo AI
Mundo AI (YC W25) builds high-quality multilingual training data for AI models, tackling the shortage of non-English data that leaves most of the world's languages underserved by today's models. The company partners with native speakers and runs end-to-end operations in the countries where target languages are spoken, using proprietary software for data collection, generation, annotation, and quality assurance. Founded in 2024 by four University of British Columbia alumni, its customers are AI labs and companies training or improving non-English and multimodal models.
Sureform
Sureform is a San Francisco startup (Y Combinator X25) that collects real-world, multimodal human data to train robots and world models. It equips workers at hotels, warehouses, restaurants, construction sites, and laundromats with head-mounted cameras, stereo depth rigs, and handheld grippers that capture time-synced video, audio, pose, depth, and tactile data while they do their jobs. Each session is delivered to robotics teams as structured, synchronized, training-ready data.
Kili Technology
Kili Technology is a Paris-based enterprise software company that helps large organizations build high-quality datasets to train, fine-tune, and evaluate AI models. Founded in 2018 by Edouard d'Archimbaud and François-Xavier Leduc, its platform spans image, video, text, PDF, and geospatial annotation, and has increasingly focused on LLM evaluation, RLHF, and expert-driven data curation for generative AI. Backed by a $25M Series A led by Balderton Capital, Kili works with enterprises including SAP, Airbus, Louis Vuitton, Michelin, and Crédit Agricole.
Superb AI
Superb AI is a computer vision AI company that builds an end-to-end MLOps and DataOps platform, letting teams label, curate, train, and deploy vision models on images, video, and 3D LiDAR data. Founded in 2018 and a Y Combinator alum, it uses AI-assisted automation to shrink the slow, manual work of preparing training data, and serves enterprise customers across autonomous driving, manufacturing, security, and public sector from offices in San Mateo and Seoul.
Halluminate
Halluminate is a San Francisco startup (Y Combinator S25) building realistic, resettable sandboxes and high-quality datasets to train and evaluate computer- and browser-use AI agents. Its flagship project, Westworld, is a fully simulated internet of synthetic consumer and enterprise apps where agents practice economically valuable tasks - booking flights, financial modeling, reorganizing data - against programmatic verifiers that produce reinforcement-learning reward signals. Customers include foundation-model labs and browser-agent companies.
AfterQuery
AfterQuery is a San Francisco applied research lab that builds expert-generated datasets, benchmarks, and reinforcement-learning environments for the world's leading AI labs. The company recruits nearly 100,000 vetted domain experts - in finance, law, medicine, software, and beyond - to teach frontier models how specialists actually think.
Labelbox
Labelbox is a San Francisco-based AI data factory that helps frontier AI labs and enterprises generate, label, and evaluate the high-quality training data their models need. Its platform combines annotation tools, model-assisted automation, and a global expert network (Alignerr) to power post-training, RLHF, and multimodal reasoning workloads.
Manu Sharma
Manu Sharma is the CEO and Co-founder of Labelbox, the leading AI data infrastructure platform that powers training data pipelines for frontier AI models. Born in Roorkee, India, he studied aerospace engineering at Embry-Riddle and Stanford before building CoPilot at DroneDeploy and leading data analytics at Planet Labs. In 2018, he co-founded Labelbox with former colleagues Brian Rieger and Daniel Rasmuson, raising $188.9M through a Series D led by SoftBank Vision Fund 2. Labelbox now serves as the command center for AI teams building and scaling production machine learning systems, from annotation and RLHF to evals and synthetic data generation.
Paroma Varma
Paroma Varma is a co-founder and Head of Solutions at Snorkel AI, the $1.3B data-centric AI platform she helped build from a Stanford AI Lab research project into a unicorn serving Fortune 500 enterprises. Holding a Ph.D. in Electrical Engineering from Stanford, she pioneered weak supervision techniques that reduce AI training data preparation from months to days, enabling domain experts in healthcare, finance, and government to build state-of-the-art ML models without massive labeled datasets.
Sama
Sama is a San Francisco-based AI training-data company that combines a 5,000-plus full-time annotation workforce - most based in Kenya, Uganda and India - with proprietary tooling to deliver labeled data, model evaluation and red-teaming for the world's largest AI builders. Founded as the nonprofit Samasource in 2008 by the late Leila Janah, it relaunched as a for-profit B Corp in 2019 and counts Google, Walmart, NVIDIA, Ford and Microsoft among its customers.
Wendy Gonzalez
Wendy Gonzalez is the CEO of Sama, a mission-driven AI data annotation and model evaluation company that employs thousands of workers in East Africa. She joined Sama in 2015, rose through the ranks as COO and President, and assumed the CEO role in 2020 following the passing of founder Leila Janah. Under her leadership, Sama closed a $70M Series B - at the time the largest funding round for a woman-led AI infrastructure company - achieved Forbes AI 50 recognition, and scaled to over 3,000 employees while maintaining its B Corp certification and social mission of lifting people out of poverty through dignified work in the AI economy.
DatologyAI
DatologyAI builds an automated data curation platform that helps companies train better AI models faster, smaller, and cheaper. Founded in 2023 by former Meta AI, DeepMind and MosaicML researchers, the Redwood City startup has raised $57.6M to industrialize the messy art of preparing petabyte-scale training data.