AI research · world models · physical systems

I build AI systems for physical worlds.

I design learning systems that reason over structure, adapt across changing conditions, and stay useful when simulation meets the constraints of the real world.

LinkedIn Google Scholar GitHub

Research Depth

Generalization beyond one simulator, dataset, or control task.

My recent research sits at the intersection of model-based RL, graph neural networks, physics-informed simulation, and engineered systems. The throughline is practical: learn useful structure from complex systems and test whether it still works when conditions shift.

Selected Research

Operator-on-F A planning-time diagnostic for latent world models. Sole-authored; RLC 2026 Workshop on World Models. The Rank-One Corner How much value equivalence does a task need from a world model? Sole-authored preprint. Bayesian Decision-Time Inference Fusing context and value priors for in-context RL. RLC 2026 Workshop on Continual RL. HOT Dataset 150,000 simulated controllable buildings for transfer research. BuildSys 2025. HVAC-SPICE In-context RL with uncertainty for zero-shot HVAC control.

Research Notes

Research writing and code releases, with a point of view.

My research Substack is now live, opening with an essay on why "how good is your world model?" is an underspecified question — writing that looks across papers rather than treating each one in isolation. More essays, code, and artifacts will follow as recent work moves from pre-publication to public release.

Beyond Research

A wider set of systems, from classrooms to sport.

Before my academic career, I competed in two Olympic Games in modern pentathlon. I have also taught and mentored in engineering and sustainability, served in governance roles, and continue to work on sustainability in sport through Racing To Zero.