I’m currently working on robotics and continual learning, particularly how agents learn through real-world experience.

A thread through much of my work is how to optimize complex systems when models are incomplete and the underlying system cannot be fully observed. Some projects:

• Most recently, I led engineering and applied science at Lithos Carbon, where we scaled enhanced rock weathering on cropland to permanently remove atmospheric CO₂. My focus was on an open problem in the field: how to reliably estimate carbon removal from sparse, delayed, and noisy measurements across heterogeneous soils.
• At Yale, I worked at QuLab on reinforcement learning for quantum error correction, learning control from partial measurements of a quantum system rather than a complete model. QuLab was led by Michel Devoret, who received the 2025 Nobel Prize in Physics for demonstrating macroscopic quantum behavior in superconducting circuits. I also worked on molecular Bose-Einstein condensates at Columbia’s Will Lab.
• At Runloop, we built simulation environments and autonomous agents for on-chain trading.

Previously, I dropped out of undergrad to start FamilyLeaf and was one of the youngest founders in Y Combinator (W12). I later joined the early team at Tilt (acquired by Airbnb), where I built Tilt Pro and led the data engineering and B2B platform teams. We were lucky to have worked with customers including Soylent, Boosted Boards, Away Travel, and the Dick’s Sporting Goods Foundation.