A local, AI-powered meeting assistant for the Transfer Tycoon project. It turns Teams meeting transcripts into structured minutes of meeting (MOMs) and lets you chat over past meetings — all on your Mac, without cloud APIs.

What it does

Piece Role
watcher.py (background agent) Watches ~/Downloads for new .docx or .vtt transcripts (e.g. from Teams). When one appears, it processes it automatically.
app.py (Streamlit UI) Chat UI at localhost where you ask questions about meetings; answers use stored summaries.

End-to-end flow

image.png

  1. Ingest — Transcript is read (.docx via docx2txt, .vtt as plain text).
  2. Generate MOM — Ollama (llama3) writes a structured MOM (project, date, attendees, agenda, discussion, action-item table).
  3. Save — MOM is saved as a Word file under ~/Desktop/MOM_Output (e.g. MoMs - [MM-DD-YY]_HHMM.docx).
  4. Index — A short, fact-dense summary is stored in ChromaDB (chroma_db/meeting_summaries) for search.
  5. Cleanup — Original transcript in Downloads is deleted after processing.
  6. Query — In the Streamlit app, your question retrieves the top 3 relevant summaries from ChromaDB, then Ollama answers using that context.

Tech stack

How to run it

You typically run both processes: