← All work
AI agents · Built solo
Agentic Meeting Assistant
A real-time assistant that listens to meetings, transcribes them on local hardware, and hands the transcript to a Claude agent that filters noise and pulls out action items while the call is still going. Other LLM agents can use it through an MCP server.
- Role
- Solo
- Transcription
- Local Whisper (GPU)
- Agent
- Claude
- Interface
- MCP server, 16 tools
Why it exists
Meetings produce commitments, and most of them get lost. I wanted the action items ready when the call ends, and I wanted other agents to be able to act on them, not just a transcript sitting in a folder.
How it works
- Capture: system audio is recorded with WASAPI loopback, so it works with any meeting app.
- Transcription: Whisper runs on a local GPU (RTX 4070 Ti), and no audio leaves the machine for a speech API.
- Agent: a Claude agent reads the live transcript, filters small talk and noise, and extracts commitments as they're made.
- MCP server: 16 tools let other LLM agents search transcripts, pull commitments, and drive follow-up work like drafting emails and creating tasks.
This is an internal system, so the diagram and write-up stand in for screenshots and code.
Let's build something.
I'm open to full-time and contract roles in AI/agent engineering, mobile, and full-stack TypeScript, remote or hybrid around NYC and Philadelphia.