An Express.js + MongoDB + TogetherAI backend that powers AI-driven chat responses, lead tracking, embeddings, and vector search for contextual knowledge retrieval.

• Powered by Meta-Llama-3.1-8B-Instruct-Turbo (TogetherAI)
• Context-aware prompt construction
• Confidence scoring:
  • Response confidence
  • Search confidence
  • Content-quality adjustments
• AI response enriched with dynamic CTAs

• Generates embeddings using BAAI/bge-base-en-v1.5 (TogetherAI)
• Stores embeddings in MongoDB Atlas Vector Search
• Falls back to regex-based text search when vector search is unavailable

• Tracks CTA clicks (with IP + GeoIP lookup)
• Stores leads in MongoDB
• Exports leads as CSV

• /api/health – Global health check
• /api/chat/health – Chat service status
• /api/track/health – Lead tracking service status

• CORS with production & development configs
• Helmet for secure HTTP headers
• Request logger (method, URL, IP, User-Agent, GeoIP location)

• Backend: Express.js
• Database: MongoDB Atlas
• AI Models: TogetherAI (LLaMA-3.1-8B, BAAI/bge-base-en-v1.5)
• Libraries: cors, helmet, dotenv, geoip-lite, request-ip

1. Clone repository

   git clone <repo-url>
   cd backend

2. Install dependencies

   npm install

3. Configure environment variables Create a .env file in the project root:

   NODE_ENV=development
   PORT=5000
   
   # MongoDB
   MONGODB_URI=mongodb+srv://<user>:<password>@cluster0.mongodb.net
   MONGODB_DATABASE=business_analysis_portal
   MONGODB_COLLECTION=knowledge_base
   LEAD_COLLECTION=leads
   
   # TogetherAI
   TOGETHERAI_API_KEY=your_api_key_here

4. Run in development

   npm run dev

5. Run in production

   npm start

• GET / → Welcome message

• GET /api/health → API health status

• POST /api/chat/message → Process chat message and return AI response
• GET /api/chat/health → Chat service health

• POST /api/track/lead → Track CTA click
• GET /api/track/export-csv → Export leads as CSV file
• GET /api/track/health → Lead tracking service health

1. User sends a message →
2. Embedding is generated →
3. Vector search in MongoDB →
4. Confidence scoring is applied →
5. AI response generated with TogetherAI →
6. Lead tracking records CTA interactions

• Structured JSON errors
• Stack trace only visible in development mode

The request logger captures:

• HTTP Method & URL
• IP Address & GeoIP location
• User-Agent