At the University of Toronto’s Computational Social Science Lab, I work closely with Dr. Ashton Anderson to lead engineering for the Maia Chess project, a research platform that studies human decision-making through chess. Maia is currently the most-played chess bot in the world, with more than four million games on Lichess and thousands of monthly users who train, analyze mistakes, drill openings, and play against it. My role is to take Maia beyond papers and training scripts and build the actual platform people interact with.

When I joined the lab, Maia existed only as a collection of papers and Lichess bots. To help Maia reach a wider audience and to foster more direct engagement with our community, I led the engineering effort to build our in-house Maia Chess platform from the ground up. This included complex frontend pipelines in TypeScript and React (Next.js)for live gameplay, move analysis, interactive drilling tools, mistake classification, user accounts, and real-time model inference.

On the applied machine learning side, I restructured Maia's inference pipeline by converting our PyTorch weights to run entirely in the browser using the ONNX web runtime. By building a client-side inference stack that runs directly on users’ machines, Maia can now serve millions of games with sub-second latency, without costing our lab a single cent for compute.