The hardest problem in robotics isn't the arm or the gripper. It's the data. Training a general-purpose robot requires a dataset that mirrors the messy, unpredictable physical world. Cortex AI, a three-person startup out of Y Combinator, is betting its $6 million seed round on building that dataset first [Y Combinator, 2025] [Preqin, 2025].
The data wedge
Cortex AI's stated mission is to build "the world's most diverse real-world, real-workplace, and industry-scale egocentric and robot datasets" [cortexrobot.ai]. Instead of starting with a robot or a specific model, the company is focusing on the foundational layer: capturing first-person video and sensor data from both humans and robots performing tasks across varied industrial and commercial environments.
The founder's pivot
Lucas Ngoo, the solo founder, brings a proven track record in scaling a consumer marketplace. As a co-founder of Carousell, he helped grow the Southeast Asian social commerce platform from an $800,000 seed round to a $35 million Series B [TechCrunch, 2013] [TechCrunch, 2016]. He stepped back from day-to-day operations at Carousell in early 2024, citing a new focus on AI [Yahoo Finance, 2024].
The scale challenge
For this bet to work, Cortex AI must solve two monumental engineering problems: collection and annotation. Gathering "industry-scale" egocentric data means deploying sensor suites across multiple customer sites, navigating logistics, privacy, and hardware reliability. The company's next twelve months will be a pure execution test: can it move from a compelling thesis to signed data-collection partnerships with real warehouses, factories, or logistics centers?