MeetScribe AI

An AI-powered meeting summarization tool that converts audio recordings into structured summaries, action items, and decisions — instantly.

🔗 https://meetscribe-ai-ten.vercel.app

---

What it does

Upload any meeting recording and MeetScribe AI will:

• Transcribe the audio using Whisper Large v3 (via Groq)
• Generate a structured summary using Llama 3.1 (via Groq)
• Extract action items, decisions, key topics, and meeting sentiment
• Store results so you can search and revisit past meetings
• Let you download a full meeting report as a .txt file

---

Tech Stack

Frontend

• React 18 + Vite
• Tailwind CSS
• React Router DOM
• Axios
• React Dropzone
• Lucide React

Backend

• Node.js + Express
• MongoDB Atlas + Mongoose
• Multer (audio file handling)
• Groq SDK (Whisper v3 + Llama 3.1-8b-instant)
• dotenv, cors, nodemon

Deployment

• Frontend → Vercel
• Backend → Render
• Database → MongoDB Atlas (free tier)

---

Project Structure

meeting-summarizer/
├── client/                        # React frontend
│   ├── src/
│   │   ├── components/
│   │   │   ├── Navbar.jsx
│   │   │   ├── UploadZone.jsx     # Drag & drop with 3 UI states
│   │   │   ├── SummaryCard.jsx    # AI analysis display
│   │   │   ├── MeetingCard.jsx    # History list item
│   │   │   └── LoadingSpinner.jsx
│   │   ├── pages/
│   │   │   ├── UploadPage.jsx     # Home — file upload + progress
│   │   │   ├── ResultsPage.jsx    # AI output after processing
│   │   │   ├── HistoryPage.jsx    # All past meetings + search
│   │   │   └── MeetingDetailPage.jsx
│   │   ├── services/
│   │   │   └── api.js             # All axios calls in one place
│   │   └── App.jsx
│   └── .env
│
└── server/                        # Express backend
    ├── config/
    │   └── groqClient.js          # Groq SDK instance
    ├── controllers/
    │   └── meetingController.js   # Request/response logic
    ├── middleware/
    │   └── upload.js              # Multer config + file validation
    ├── models/
    │   └── Meeting.js             # Mongoose schema
    ├── routes/
    │   └── meetingRoutes.js       # API route definitions
    ├── services/
    │   └── meetingService.js      # Whisper + Llama 3 pipeline
    ├── uploads/                   # Temporary audio storage
    └── index.js                   # Entry point

---

Getting Started

Prerequisites

• Node.js 18+
• A free Groq API key
• A free MongoDB Atlas cluster

1. Clone the repo

git clone https://github.com/yourusername/meetscribe-ai.git
cd meetscribe-ai

2. Set up the backend

cd server
npm install

Create server/.env:

PORT=5000
MONGO_URI=your_mongodb_atlas_connection_string
GROQ_API_KEY=your_groq_api_key

Start the server:

npm run dev

Server runs on http://localhost:5000. Verify at http://localhost:5000/health.

3. Set up the frontend

cd client
npm install

Create client/.env:

VITE_API_URL=http://localhost:5000/api

Start the frontend:

npm run dev

App runs on http://localhost:5173.

---

API Endpoints

Method	Endpoint	Description
GET	/health	Server health check
POST	/api/meetings/upload	Upload and process audio file
GET	/api/meetings	Get all completed meetings
GET	/api/meetings/:id	Get one meeting by ID

Upload request

POST /api/meetings/upload
Content-Type: multipart/form-data

Fields:
  audio  (File)    — required, max 25MB, audio formats only
  title  (String)  — optional

Upload response

{
  "message": "Meeting processed successfully",
  "meeting": {
    "_id": "64abc...",
    "title": "Q3 Sprint Planning",
    "transcript": "Alright everyone, let's get started...",
    "analysis": {
      "summary": "The team aligned on Q3 release scope and assigned QA ownership.",
      "actionItems": ["Ravi to update test plan by Friday"],
      "decisions": ["Release pushed to August 15th"],
      "sentiment": "positive",
      "keyTopics": ["release planning", "QA", "deadlines"],
      "duration": "approximately 25 minutes"
    },
    "status": "completed",
    "fileSize": "3.12 MB",
    "createdAt": "2025-01-15T10:30:00.000Z"
  }
}

---

AI Pipeline

Audio file (mp3/wav/webm)
        ↓
  Groq Whisper Large v3
        ↓
  Raw transcript text
        ↓
  Groq Llama 3.1-8b-instant
  (structured prompt → JSON)
        ↓
  { summary, actionItems, decisions,
    sentiment, keyTopics, duration }
        ↓
  Saved to MongoDB Atlas

Why Groq? Groq's free tier includes both Whisper and Llama 3.1 with fast LPU-based inference — no credit card required. Ideal for a production-grade prototype.

Reliable JSON extraction — LLMs occasionally wrap responses in natural language even when instructed not to. A regex fallback parser extracts the JSON object regardless:

try {
  return JSON.parse(raw)
} catch (e) {
  const match = raw.match(/\{[\s\S]*\}/)
  if (match) return JSON.parse(match[0])
  throw new Error('Could not parse AI response')
}

---

Deployment

Backend → Render

1. Push server/ to GitHub
2. Create a new Web Service on Render
3. Set build command: npm install
4. Set start command: node index.js
5. Add environment variables: PORT, MONGO_URI, GROQ_API_KEY

Frontend → Vercel

1. Push client/ to GitHub
2. Import the repo on Vercel
3. Add environment variable: VITE_API_URL=https://your-render-url.onrender.com/api
4. Deploy

Note: Render's free tier spins down after inactivity. The first request after idle may take ~30 seconds to wake up — expected behaviour for a portfolio project.

---

Supported Audio Formats

Format	Extension
MP3	.mp3
WAV	.wav
WebM	.webm
MPEG-4 Audio	.m4a
OGG	.ogg

Maximum file size: 25MB (Groq API limit)

---

Environment Variables

Server

Variable	Description
PORT	Server port (default: 5000)
MONGO_URI	MongoDB Atlas connection string
GROQ_API_KEY	Groq API key from console.groq.com

Client

Variable	Description
VITE_API_URL	Backend base URL

---

Future Improvements

• JWT authentication — scope meetings per user
• Job queue (BullMQ) — async processing so HTTP requests don't hang
• S3/R2 storage — replace ephemeral local uploads folder
• Streaming responses — show transcript as it's generated
• Speaker diarization — identify who said what
• Export to PDF/Notion/Slack

---

License

MIT