🎙️ Meeting Intelligence System

Upload a meeting recording → Get transcript, summary, and action items in under 60 seconds.

---

📸 Demo

Upload & Auto-route	Transcription + Summary

Meeting Summary	Action Items

---

🚀 Quick Start

git clone https://github.com/ShainaHussain/Meeting-Intelligence-System.git
cd Meeting-Intelligence-System
pip install -r requirements.txt

# Add your API key
echo "GROQ_API_KEY=your_key_here" > .env

streamlit run app.py

Get a free Groq API key at console.groq.com

---

✅ What It Does

Upload any meeting recording and get back:

• 📊 AI Summary — 3-5 sentence overview of the entire meeting
• ✅ Action Items — structured list with owner, task, and deadline
• 📝 Full Transcript — complete text of everything said
• 💾 Meeting History — all past meetings saved locally
• 📥 Download Report — full report as a .txt file

---

⚡ How Fast

Meeting Length	Processing Time
10 minutes	~30 seconds
30 minutes	~1 minute
1 hour	~2-3 minutes

---

🧠 How It Works

Upload Audio
     ↓
Smart Routing (file size check)
     ├── < 25MB  → Groq Whisper API  (fast, free)
     └── > 25MB  → AssemblyAI        (up to 5GB)
     ↓
Validation (word count + keyword check)
     ↓
LLM Intelligence — Llama 3.3 70B via Groq
     ├── Meeting Summary
     └── Action Item Extraction (chunked for long meetings)
     ↓
SQLite DB → Save meeting record
     ↓
Display + Download

Key design decisions:

• Hybrid transcription — auto-routes between Groq and AssemblyAI based on file size. No manual selection needed.
• Chunking strategy — splits transcripts into 5000-word chunks to handle token limits. Works for 3+ hour meetings.
• Lazy client init — Groq client created only when a key exists, preventing startup crashes on cloud deployment.
• 3-layer secret management — checks st.secrets → .env → sidebar input. Same code works locally and in production.

---

🛠️ Tech Stack

Layer	Technology	Why
UI	Streamlit	Rapid deployment, no frontend code needed
Transcription (small)	Groq Whisper API	Free, fast, state-of-the-art
Transcription (large)	AssemblyAI	Handles async, up to 5GB
LLM	Llama 3.3 70B via Groq	Free tier, fast inference
Database	SQLite	Zero setup, built into Python
Deployment	Streamlit Cloud	Native support, free tier

---

💰 Cost

Completely free for most use cases.

• Groq API — free tier handles ~100 meetings/day
• AssemblyAI — 5 hours/month free (for large files only)
• Hosting — Streamlit Cloud free tier

---

📁 Project Structure

Meeting-Intelligence-System/
├── app.py              # Main application (transcription + LLM + UI)
├── requirements.txt    # Dependencies
├── learning_log.md     # Day-by-day build journal
├── .gitignore
└── README.md

---

🗺️ Roadmap

Completed ✅

• Hybrid transcription (Groq + AssemblyAI auto-routing)
• AI meeting summary
• Action item extraction with chunking
• 95+ language support with auto-translation
• SQLite meeting history
• Download complete report

In Progress 🔄

• Speaker diarization (who said what)
• PDF report export

Planned 📋

• Persistent cloud database (Supabase)
• Email report to yourself
• Search across past meetings
• FastAPI wrapper for programmatic access

---

⚠️ Known Limitations

• SQLite resets on server restart (Streamlit Cloud) — use locally for persistent history
• Speaker diarization requires AssemblyAI key (not available via Groq)
• No real-time transcription — file upload only

---

📖 Learning Log

This project was built incrementally over several days. See learning_log.md for a day-by-day breakdown of what was built, problems hit, and how they were solved. A real record of the engineering process.

---

License

MIT — free to use and modify.

---

Built by Shaina Hussain
LinkedIn · LeetCode · Email