Local-first LLM agents, MCP tooling and automation, taken from notebook to something that actually runs.
Click any card to open the repository.
A fully local voice agent. faster-whisper handles speech-to-text, Qwen 2.5 3B (via Ollama) reasons, and a FastMCP server exposes tools over SSE. VAD, Piper TTS and a WebSocket UI are next.
An AI agent that finds new jobs matching your requirements and automates the application flow.
FastAPI backend that ingests trades across venues, tracks positions with weighted-average cost, and computes realized and unrealized PnL with strict idempotency.
Interactive simulator that visualizes how packets flow hop by hop across a network topology.
A simple, fast password generator anyone can use. Pick the rules, get a strong password, copy it.
Real-time chat application built with Socket.io.
Three small interactive demos. Everything runs in your browser.
The architecture of my VoiceAI Agent. Stages marked built run today; planned ones are next on the roadmap.
Mirrors the tools on my FastMCP server: get_current_time, calculate and get_weather. Here a simple router stands in for the Qwen 2.5 3B model that picks the tool in the real agent.
Type any text. See it split into tokens, then placed on a 2D map of meaning next to related phrases. This is a teaching sketch: a simplified tokenizer and a keyword-based map, not a real model.
A small assistant grounded in my resume and projects.
Like a training run: the loss keeps going down as you scroll.