I work on reinforcement learning for frontier model post-training and robotics.
I led an RL research team, and our algorithms reshaped post-training for every model release from GPT-5 onwards.
I helped kickstart OpenAI's robotics revival in 2024, training our first vision-language-action models.
We open-sourced CeLLM: drag one formula to apply a prompt or example outputs across a whole column.
Made at Chroma's hacker-in-residence, CLAIRe learns the people and places in your life and checks in on you.
PaVE combines QB selection, completion odds, and pass value across up to 1.3 million passes per game-second.
I shared a photo of my childhood home, hoop by the driveway, and a model nailed the ZIP code from it.
I work on reinforcement learning for frontier model post-training and robotics.
I led an RL research team, and our algorithms reshaped post-training for every model release from GPT-5 onwards.
I helped kickstart OpenAI's robotics revival in 2024, training our first vision-language-action models.
We open-sourced CeLLM: drag one formula to apply a prompt or example outputs across a whole column.
Made at Chroma's hacker-in-residence, CLAIRe learns the people and places in your life and checks in on you.
PaVE combines QB selection, completion odds, and pass value across up to 1.3 million passes per game-second.
I shared a photo of my childhood home, hoop by the driveway, and a model nailed the ZIP code from it.
By using this service, you agree to the Terms of Service and Privacy Policy.