Reinforcement Learning and Flow Matching for End-to-End Driving
2026Driving policies trained end to end, combining reinforcement learning with diffusion and flow-matching policy classes.
Scholar GitHub Twitter LinkedIn
I study Engineering Science at the University of Toronto, and work as a software engineer at Tenstorrent on distributed ML and low-latency inference.
I am a student researcher with the University of Toronto and NVIDIA Research, working on end-to-end driving, reinforcement learning, and diffusion and flow-matching models. Previously I was a machine learning intern at Ode, with Clay and LGND AI, finetuning geospatial foundation models.
My interests are in applications of AI to the physical world — geospatial machine learning and autonomous vehicles in particular.
Courses I've Taken in University · A Collection of Cool Companies (to me) · International Geography Olympiad 2023
Machine Learning · Reinforcement Learning · ML Systems · Robotics · Math · Algorithms
A linked, searchable notebook — browse all of it
End-to-end driving with RL and flow matching · RealADSim challenge, 1st place · Geospatial foundation models
Scroll or click to expand
Driving policies trained end to end, combining reinforcement learning with diffusion and flow-matching policy classes.
Our winning entry to the closed-loop driving challenge, evaluated in a reconstructed-real-world simulator rather than on logged trajectories.
Adapted open geospatial foundation models to downstream Earth-observation tasks, and built the evaluation around them.
AlphaEarth Foundations · Predicting food deserts · Solar PV forecasting · IGeo
Scroll or click to expand
A from-scratch implementation of DeepMind's embedding model for Earth observation.
Represented Canada at the International Geography Olympiad.