Projects

Reinforcement learning agent targeting >96% win rate on Slay the Spire Watcher at Ascension 20 — the hardest difficulty, where top players achieve ~90%. Built a complete game engine with 100% Java parity, a Rust combat engine, and a PPO training pipeline with a 3M-parameter StrategicNet.

AI Engineer at IBM building AI tooling for Mainframe Code Assistant - enabling mainframe code understanding and translation using LLMs.

Sunrise alarm clock TUI to gently ease one into the world.

Data Scientist role at Royal Bank of Canada, implementing advanced time series forecasting.

Research investigating how model quantization affects sparse autoencoders (SAEs). Found SAEs transfer across precisions (99% correlation), code generation degrades 50% at INT4 while knowledge retrieval survives, and undercomplete SAEs (fewer features than model dimension) transfer 2.3x better than overcomplete ones.

Interactive solvers for New York Times word puzzle games with web UI.

Hackathon project forecasting linguistic patterns to predict how language evolves over time.

Interactive website for teaching linear algebra fundamentals, built while TAing at Queen's University. Used as primary resource for 80+ students across 2 tutorial sections.

A multi-agent simulation system for optimizing garbage collection drone operations in the Gulf of Mexico.

Reverse-engineered unknown dynamic systems through experimental identification, designed stable PID controllers.

Led a team developing trading strategies using neural networks and portfolio optimization.

Built options pricing models using Monte Carlo simulation and the Heston model.

Data Analyst role at Royal Bank of Canada, focusing on automation and forecasting.