Rajat Bhageria is the Founder and CEO of Chef Robotics, a physical AI company that automates food production. Previously, he was a Founder and Managing Partner at Prototype Capital, a pre-seed venture capital fund investing in founders who apply new technology to old industries. Before Prototype Capital, Rajat founded ThirdEye, a company that developed assistive technology for the visually impaired. ThirdEye was ultimately acquired. Rajat holds a Master’s degree in Robotics and Machine Learning and a Bachelor’s degree in Economics from the University of Pennsylvania.
Physical AI has made remarkable strides, but most foundation models are trained on rigid objects. Food represents one of the most complex manipulation challenges in robotics: every ingredient is deformable, inconsistent in weight and texture, sensitive to temperature, and must be handled with calibrated force across thousands of variations. Chef Robotics has built the largest real-world dataset of deformable material manipulation by completing over 118 million food servings in production across over a dozen food manufacturing facilities in North America and Europe. This data serves as the basis for our Food Foundation Model (FFM), which enables robots to generalize to new ingredients with minimal retraining. This session will explore what makes food such a demanding benchmark for physical AI: the sensor-fusion challenges of grasping soft and unpredictable objects, the force-control precision required to handle fragile versus dense materials, and why real-world variation at scale is irreplaceable for training robust policies. More importantly, we’ll show why solving food unlocks progress far beyond this industry. The techniques developed here (e.g., adaptive grasping, tactile feedback integration, high-variance training distributions) may transfer to medical devices, flexible packaging, agriculture, and other domains involving deformable materials. Attendees will leave with a concrete framework for thinking about deformable material manipulation as the next frontier in physical AI, and evidence that real-world data at scale is what separates lab demos from deployable systems.
