I build agentic AI systems that hold up in practice.

I build agentic AI platforms, assistants, and ML systems with one bias: they should be observable, evaluable, and useful after the demo.

My research started with bandits, safe best-arm identification, Bayesian optimization, and out-of-distribution detection. Today I am focused on multi-agent reinforcement learning in asynchronous environments, LLM evaluation, and the infrastructure needed to make AI systems measurable and robust.