A from-scratch LLM inference engine (zero runtime dependencies, verified token-for-token against llama.cpp) and a browser lab that shows every stage of next-token prediction on real weights: watch attention read, the logit lens form an answer layer by layer, then fork history and compare both futures. Runs entirely in your browser.

Spaced-repetition flashcard app for Hiragana and Katakana. ML-powered handwriting recognition grades stroke accuracy, while an adaptive SRS algorithm schedules reviews based on your performance.

Platform for the #1 best-selling Japanese language learning book series on Amazon. Rebuilt from a legacy PHP codebase to a modern MERN stack — React front-end, Express API, and MongoDB-powered dynamic content throughout.

CS5001 final project — a flood-fill puzzle with a live recursive call stack visualizer. Watch the algorithm's stack frames update in real time as you play. Built with Python and Pygame, compiled to WebAssembly via Pygbag.