AI Meeting Summarizer & Action Tracker

Live Demo URL: https://meeting-summarizer-735199411680.europe-west3.run.app/

A hackathon-friendly project that takes in meeting transcripts and produces:

• A concise summary
• Key decisions
• Structured action items (assignee (if mentioned), task, due_date (if available) and other_notes (if mentioned))

Features

• Upload files .txt, .pdf, .docx or paste text directly
• Supports OCR for scanned PDFs
• Supports audio transcription using Faster-Whisper
• Local data persistence for meeting history

Tech

• Meta LLaMA (model family) served via Cerebras (fast inference)
• Python 3.11 + Streamlit UI
• SQLite for simple persistence
• Docker for portable deployment
• Google Cloud Run

Quick Start (Local)

1. Create a virtualenv and install deps:

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
brew install poppler tesseract ffmpeg   # for OCR + audio decoding

3. Run the app:

streamlit run app/main.py

If Cerebras is not set up, the app will fall back to a lightweight mock so you can test the UI end-to-end.

Environment

• LLM_PROVIDER: one of cerebras, mock
• CEREBRAS_API_BASE: base URL for Cerebras (e.g., https://api.cerebras.ai/v1) — adjust per docs
• CEREBRAS_API_KEY: your key/tokens

Docker

Build & run:

docker build -t meeting-summarizer .
docker run -p 8501:8501 --env-file .env meeting-summarizer

Notes

• The cerebras client is implemented assuming /chat/completions API schema. Adjust fields per your actual Cerebras endpoint docs if needed.