Injects an LLM-powered academic advisor widget into the Tufts student portal by intercepting encrypted HTTPS traffic through a custom C mitm-proxy server. The proxy has a non-blocking `libev` based architecture allowing for simultaneous connections, and spoofs TLS certificates as a CA to read and modify the encrypted traffic. Intercepted pages are sent over Unix sockets to a Python server that injects the widget's JavaScript and handles backend requests. A *gpt-4.1-mini* agent answers student questions using RAG queries against a database of structured Tufts course summaries scraped from official sources. Uploaded PDF transcripts are parsed into structured JSON (e.g. GPA, major, lists of courses taken) using a seperate LLM and appended to the system prompt, while info from the users current page is parsed from the intercepted HTML using BeautifulSoup and included in each message for contextually relevant responses. Containerized with Docker for portable deployment.

A wireless MIDI accordion built around dual ESP32 microcontrollers communicating via ESP-NOW. An Arduino Mega handles analog reads from 8 keys with soft potentiometers and FSRs, while the ESP32 processes and transmits TOF-based bellows data to a Max/MSP patch for MIDI conversion. Self-contained with battery power and laser-cut acrylic construction.

A performant grid-based particle simulation modeling the behavior of sand, water, stone, and gas elements with physics interactions. Uses the RayLib C Graphics Library. Built using a modular OOP architecture with base classes for particle types and a centralized world grid managing the simulation loop.

A Guitar Hero inspired rhythm game implemented entirely in SystemVerilog on an iCE40 FPGA. Features VGA graphics output, buffered I2S audio playback from a MicroSD, realtime strum-gated note registration. Uses custom python tools for MIDI to note data conversion and direct writes of processed audio data to MicroSD card sectors.

A robot vacuum simulator controlled by a hierarchical behavior tree AI. Features dynamic dust accumulation, battery management with priority-based charging, spot and general cleaning modes, and a real-time GUI visualization of the discrete grid environment.