I work at Google on Post Training Gemini. I was a core contributor to Gemini 3.0, 2.5, and 2.0, where my work helped push our model performance to the top of the WebDev Arena and SWE benchmarks.

I’m interested in how we get models to discover and solve new problems that are unknown to us. My current focus is on long-horizon RL and building reliable reward models to guide that exploration.

I also like to write about the quiet ways AI is changing how we think.