Developed in collaboration with the New Jersey Department of Health, this Android app identifies six major malaria-causing parasites from microscopic blood smear images. Detected parasites are shown with bounding-boxes and confidence scores, giving healthcare workers quick, interpretable results. I trained the deep learning model powering the app, and optimized it to run directly on-device for detection even in low-resource settings.

A full-stack web app that generates personalized, shareable reading summaries from Goodreads data--think Spotify Wrapped, but for books. Built because Goodreads' own year-end review felt lacking, this project analyzes uploads to uncover patterns in pages read, top authors, and monthly trends, with interactive visualizations designed for friends to explore and share.

Two three-layer neural networks built entirely from scratch in NumPy--no external ML libraries. One classifies handwritten digits (MNIST), including hand-drawn user input; the other distinguishes faces from non-faces. Building these deepened my understanding of how neural networks learn from data.

An OS-inspired digital playground built to capture the nostalgia of first discovering a computer. Designed around intentional constraints and micro-interactions, it favors reflection over productivity. At launch, the site resonated widely, drawing more than 2,000 visitors who explored its hidden routes and playful widgets.