voice AI · knowledge graphs · MIDI hardware
A production-minded proof of concept for FastRTC — real-time voice mode in Python. FastAPI + React/TS backend and frontend, ElevenLabs and Silero VAD, with detailed docs on tuning audio for low latency.
The production ops layer FastMCP doesn't ship with: API key issuance, rotation and revocation, per-key and per-tool rate limiting, and pluggable audit logging (file, SQLite, HTTP, OpenTelemetry). Published on PyPI.
LLM agents that extract and maintain knowledge graphs from text and live conversation — one POC pulls structured entities out of raw text, the other keeps a running graph of a user updated as a conversation unfolds.
A client-server AI chatbot built for academic institutions — students chat with an AI tutor grounded in course material via RAG, faculty upload docs, admins manage courses. The basis for a first-author IEEE paper.
A reusable React component library for dropping WebRTC voice chat into any app, built on top of FastRTC.
A web app for worship teams to organize, manage, and share songs — chord charts, lyrics, and setlists in one place.
A sheet-music library so I can pull up any chart from anywhere, instead of hauling around a binder.
A 29-note MIDI pedalboard for a pipe organ, built from two Arduino Leonardos wired up as native USB-MIDI devices — no extra software needed, just plug in and play through Hauptwerk.
A DIY 12-button, 5-potentiometer USB MIDI controller built with Arduino, for live rigs and Mainstage.