I work at the intersection of three things: teaching AI to act in the physical world, learning to trust what agents decide, and carrying the rigor of science into both.
Physical AI
Embodied intelligence that has to work in the real world — robot manipulation, sim-to-real, and large-scale simulation data. From LLM-driven grasping to distributed synthetic-data pipelines.
Agent Evaluation
The softest spot in AI agents: why should anyone trust a multi-agent system’s decisions? I bring uncertainty quantification and honest negative results to how we measure agents.
AI for Science / UQ
The rigor of earth science — uncertainty quantification, simulation, geostatistics — carried into AI. Turning scientific methodology into communicable ideas and sharper models.
→ Read the writing, follow the streams, see the projects, or learn more about me.