Building physics-grounded environments and evaluations for physical AI.
Founder @ Second Nature Labs
I’m the founder of Second Nature Labs, where we build physics-grounded RL environments and evaluations for AI systems that need to reason about and act within the physical world.
Our work centers on a simple question: when an agent encounters a real-world system, can it diagnose what is actually happening, choose the right intervention, and verify that the problem is solved?
I’m interested in environments where success requires causal reasoning, disciplined action, and measurable physical correctness rather than pattern matching alone.
Currently:
- Founder @ Second Nature Labs
- Duke University Class of 2026 — Mechanical Engineering, Computer Science, Innovation & Entrepreneurship
- Building open benchmarks for agentic and physical AI
- Physics-grounded simulation environments
- Agentic reinforcement learning and evaluation
- Real-world diagnostic reasoning
- Cheat-resistant graders and benchmarks
- Infrastructure for robotics and physical AI teams
An Inspect-native agent environment and benchmark where language models diagnose simulated vehicle electrical no-start and charging faults.
Agents must gather evidence, identify the root cause, make disciplined repair decisions, and verify the fix through a successful start. The environment uses computed electrical physics and evaluates performance across diagnostic accuracy, parts discipline, and time efficiency.
Beyond physical AI, I’ve worked on projects spanning:
- Quantitative modeling and portfolio optimization
- Financial markets and forecasting
- Applied machine learning
- Decision-making under uncertainty
I’m always interested in conversations about physical AI, agent evaluation, simulation, and the infrastructure needed to make intelligent systems dependable in the real world.