A simple yet powerful Retrieval-Augmented Generation (RAG) system built with Node.js, Groq API, and local embeddings. Upload PDF documents and ask questions about their content with real-time streaming responses.

• PDF Upload & Processing: Upload multiple PDF documents (up to 50MB each)
• AI-Powered Q&A: Ask questions and get intelligent answers based on your documents
• Smart Search: Uses vector embeddings to find relevant information
• Source Citations: See which parts of your documents were used in answers
• Real-time Streaming: Watch responses generate in real-time
• Cost-Effective: Free local embeddings + Groq's fast, affordable LLM
• Clean UI: Modern, intuitive web interface

• Backend: Node.js + Express + Socket.IO
• Frontend: HTML5 + CSS3 + Vanilla JavaScript
• PDF Parsing: pdf-parse
• Embeddings: @xenova/transformers (local, free)
• Vector Storage: In-memory with cosine similarity
• LLM: Groq API (openai/gpt-oss-20b)

• Node.js 16+ installed
• Groq API key (get one at console.groq.com)

1. Clone or download this project

2. Install dependencies:

   npm install

3. Set up environment variables:

   • Edit .env file
   • Add your Groq API key:

     GROQ_API_KEY=your_actual_groq_api_key_here
     PORT=3000
     

4. Start the server:

   npm start

5. Open your browser:

   • Navigate to http://localhost:3000

1. Click the upload area or drag & drop a PDF file
2. Click "Upload & Process"
3. Wait for processing (first-time embedding model download may take a moment)
4. Your PDF is now ready for queries!

1. Type your question in the chat input
2. Press Enter or click Send
3. Watch the response stream in real-time
4. See source citations at the bottom of each answer

• View document statistics in the sidebar
• Clear all data with "Clear All Data" button
• Upload multiple PDFs to expand your knowledge base

Edit src/ragService.js to adjust:

• chunkSize: 1000 (characters per chunk)
• chunkOverlap: 200 (overlap between chunks)
• topK: 3 (number of chunks to retrieve)

Edit server.js to change:

• Groq model: openai/gpt-oss-20b
• Temperature: 0.7 (higher = more creative)
• Max tokens: 2048

groq-rag-system/
├── package.json          # Dependencies
├── .env                  # Environment variables
├── server.js            # Express server + Socket.IO
├── src/
│   ├── pdfService.js     # PDF text extraction
│   ├── embeddingService.js # Local embeddings
│   ├── vectorStore.js    # In-memory vector DB
│   └── ragService.js     # RAG orchestration
└── public/
    ├── index.html        # Frontend UI
    ├── style.css         # Styling
    └── script.js         # Client-side logic

1. Upload PDF: Extract text from uploaded PDF
2. Chunk Text: Split text into smaller, manageable chunks
3. Generate Embeddings: Convert chunks to vector embeddings (local, free)
4. Store Vectors: Save embeddings in memory with metadata
5. Query Process:
   • Convert question to embedding
   • Find similar chunks using cosine similarity
   • Retrieve top-k relevant chunks
   • Send chunks + question to Groq LLM
   • Stream response back to user