Selected work
A few of my projects — on-device AI, developer tools, and scientific computing.
On-device AI
Browser extension
Grammar and spelling corrections powered by small language models running on your device. Your text stays local.
Qwen models run locally through Transformers.js with WebGPU or WASM, with Gemini Nano as an optional engine. Sentence caching avoids repeated checks, while word-level diffs place suggestions in the right text range.
Developer tools
Web application
Everyday developer utilities in one browser-based workspace: format JSON, inspect JWTs, compare text, and work with PDFs and images.
Angular progressive web app with lazy-loaded tools and local file processing. A shared WebLLM worker handles on-device AI without blocking the interface.
Scientific computing
CERN-HSF · GSoC 2019
A Jupyter Notebook extension for managing Python dependencies per SWAN project, designed to make shared scientific notebooks easier to reproduce.
Python server APIs and a JavaScript notebook interface, with a separate conda environment for each SWAN project. The prototype keeps project dependencies separate.