Independently reproduced and extended a state-of-the-art deep learning model for 3D spatial audio localization, achieving 99% detection accuracy and a 0.9° azimuth error across 3 trained room environments. Demonstrated strong generalization with 94% accuracy on completely unseen acoustic environments, validated through 3-fold cross-validation.

A fully local RAG-powered research assistant in Rust with an MCP-compatible interface, enabling LLMs to semantically search personal document collections via natural language. Features a high-performance vector store with automatic file indexing, PDF extraction, and persistent FAISS-backed similarity search — no cloud dependencies.

A computer vision targeting system combining real-time object tracking with embedded ESP32 motor and pneumatic control, packaged in a custom-designed 3D-printed chassis.

An AI drawing game integrating real-time computer vision, a custom TensorFlow gesture classifier, and a Gemini API judging system. Won "Best Use of Gemini API" at Bitcamp 2025 out of 90+ teams and was selected to demo to engineers and project managers at Google DeepMind.

A personal portfolio site built with Next.js, React, and Node.js. It showcases my background and projects using a clean, modern layout for simple navigation and easy updates.

A web app to automate the process of organizing our yearly Secret Santa gift exchange. It allowed users to register and pick gifts to be used during the party.

A Rust program implementing reference counting, mark-and-sweep, and stop-and-copy garbage collection algorithms. Contact me if you'd like access to audit the code (non-students only).

A dynamically-typed type-checked subset of OCaml, including a lexer, parser, and interpreter (mutop). Contact me if you'd like access to audit the code (non-students only).

A full-stack Flask web app to help students navigate university course catalogs using natural language. Uses regex to extract data from PDF catalogs into structured JSON, then converts natural language queries into SQL. Built in 48 hours at BitCamp 2024 — received an honorary mention from event sponsors.

A PyTorch + OpenCV model for distinguishing between four wooden cross-sectional topographies. Used in ENES100 Fire Suppression task.

A Python script leveraging OpenCV's image processing to isolate shapes within cards. Implemented contour approximation to automate gameplay for the board game 'Spot It!'. Achieved approximately 76% accuracy.

A video game where players defeat enemies by typing correct arrow key combinations. Includes a dynamic leveling system with increasing enemy difficulty.

A custom Linux desktop with several packages and tailored configurations.