About us
A thousand picks is roughly one shift on one line. Each mark is one pick and each coral mark is a stoppage. At 85% per attempt the line stops about 150 times. At 99.9% it stops once.
Generality is the axis the field is optimising and it is not the axis a plant buys on. A policy that attempts a thousand different tasks and succeeds at most of them is a research result. A cell that does one job every time is a position on a shift schedule, and the difference between those two things is the last few points of reliability, which do not come from scale. They come from failing in a real building and fixing it.
So we picked one job: taking the part out of the bin, building the kit, bringing it to the line. The policy runs in simulation today, and we are in conversation with two factories about putting the first cell on their floor.
Co-founder, CEO
Reinforcement learning research, and sim-to-real transfer closed first-hand. He built OpenDOG, an open source RL quadruped trained entirely in simulation and deployed onto 3D-printed hardware. First nationally at HackDef 2025 and second in the Latin American Cybersecurity Challenge 2025. Founded GENESIS, Mexico's largest university aerospace team. Started reverse engineering games at nine.
Co-founder, CTO
Hardware that shipped. He took Sunu from prototype to a product in 50 countries, past $1M in sales and 5,000 users. Twice a YC founder, with Sunu in S17 and Clara in P26. Inventor on two patents, Techstars Miami 2023, and twice on Shark Tank.
Verified founder Sunu S17 · Clara P26