Cortex AI is building the worlds most diverse large-scale dataset of real-world workplace robot and egocentric human data to train and evaluate embodied AI models. The thesis is that lab demonstrations and pure simulation only get robotics foundation models so far, and that the next leap in physical intelligence requires data captured in the messy, unpredictable environments where robots will actually have to work. Cortex AI was founded in 2025 by Lucas Ngoo, previously co-founder of the billion dollar consumer marketplace Carousell, and is based in San Francisco with three employees as part of the Y Combinator F25 batch.
The platform delivers three data products. Egocentric Data provides first-person workplace video annotated with hand and body pose, depth, and subtask labels. Robot Data captures trajectories from real manipulators and humanoids deployed inside live industrial settings. Human-in-the-Loop Rollouts and Evals support real-world deployments with remote operators who recover robots when they fail, producing both edge-case training data and rigorous evaluation signal. The Cortex Marketplace pays workplaces to host these data collection and evaluation sessions while giving labs access to in-the-wild data that simulations cannot reproduce. The company has already enabled training of the MolmoAct2 robot foundation model.
Cortex AI should not be confused with the existing Cortex.io developer platform, the Cortex XSOAR security automation product, or any of the other companies using the Cortex name. This Cortex AI is exclusively focused on data infrastructure for embodied AI and humanoid robotics. Customers are robotics labs, humanoid companies, and foundation model teams that need real-workplace data to push policies beyond the lab. Pricing is contact-sales, with engagements structured around dataset licenses, custom collections, and ongoing human-in-the-loop evaluation contracts.