How shynbui/meeting-minutes-generator works
Project Type and Purpose
• This is a lightweight full-stack application for generating meeting minutes from a transcript.
• It consists of a Python FastAPI backend (app/) that handles core logic (preprocessing, summarization via OpenAI/Gemini, export to Word) and a Gradio frontend (ui/) for user interaction.
File Structure Analysis
• Top-level:
• .env – environment variables for API keys
• run.py – script to spin up both FastAPI and Gradio (optional)
• requirements.txt – Python dependencies
• app/: FastAPI service
• main.py – FastAPI entry point
• config.py – global configuration (API keys, model parameters)
• modules/
– preprocessing.py
– summarizer.py
– exporter.py
– schema.py
• ui/: Gradio interface
• interface.py – launches web UI, takes user transcript, displays/downloads minutes
a. User Interface (Gradio)
b. API Layer (FastAPI)
c. Processing Modules (chain of internal services)
• Preprocessor
• Summarizer (calls external LLM API)
• Exporter (writes .docx via python-docx)
d. External Services
• OpenAI or Gemini API
e. Configuration/Environment (.env)
Relationships & Data Flow
1. User uploads/transcribes meeting via Gradio UI
2. UI sends transcript to FastAPI endpoint (“/generate”) over HTTP
3. FastAPI/main.py loads config.py, then orchestrates modules in sequence:
• preprocessing → summarization (external API call) → formatting/schema → exporter
4. Exporter returns a .docx file path or binary
5. FastAPI responds to UI with generated document
6. UI displays download link for the meeting minutes
Architectural Patterns & Design Principles
• Layered Architecture: UI → API → Business Logic modules → External API
• Single-responsibility modules (each does one step: preprocess, summarize, export)
• Configuration-driven (via config.py and .env)
• Stateless HTTP service; no persistent database
Essential Details to Capture
• File-generation step and delivery back to the UI.