DoctorN8/ABK_Learning_Solutions

💬 Website RAG Chatbot

A powerful AI-powered chatbot for your website that uses RAG (Retrieval-Augmented Generation) to answer questions about your services, projects, and content. Built with modern NLP techniques, FAISS vector search, and LLM integration.

✨ Features

• —🤖 LLM-Enhanced Responses: Uses Groq (fast) or Ollama (local) for intelligent answers
• —🔍 RAG Architecture: Combines semantic search with LLM reasoning
• —🌐 Web Interface: Beautiful Gradio UI for easy interaction
• —📚 Source Citations: All answers include specific source references
• —⚡ Fast Semantic Search: FAISS-powered vector search
• —🔧 Multiple LLM Providers: Choose between Groq (API) or Ollama (local)
• —📝 Markdown-Based: Simply add your website content as markdown files

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Configure API Key

# Copy the example environment file
copy .env.example .env

# Edit .env and add your Groq API key
# Get free key at: https://console.groq.com

Edit .env:

GROQ_API_KEY=your_actual_api_key_here
LLM_PROVIDER=groq
LLM_MODEL=llama-3.3-70b-versatile

3. Add Your Content

Place your markdown files in data/markdown_files/ directory. The system will automatically:

• —Load all .md files
• —Create embeddings
• —Build the search index

4. Launch the Chatbot

python start_web.py

Open your browser to: http://localhost:7860

📁 Project Structure

Website RAG Chatbot/
│
├── data/
│   └── markdown_files/          # Your website content (markdown files)
│       ├── services.md
│       ├── about.md
│       ├── contact.md
│       └── ... (add your 19 files here)
│
├── src/
│   ├── data_loader.py           # Loads markdown files
│   ├── embedding_indexer.py     # Creates FAISS index
│   ├── llm_generator.py         # LLM response generation
│   ├── search.py                # Unified search interface
│   └── main.py                  # Gradio web interface
│
├── app.py                       # Entry point for deployment
├── start_web.py                 # Local startup script
├── requirements.txt             # Python dependencies
├── .env.example                 # Environment variables template
└── README.md                    # This file

🔧 How It Works

1. 1.Data Loading: Markdown files are loaded from data/markdown_files/
2. 2.Embedding Creation: Text is chunked and converted to embeddings using sentence-transformers
3. 3.Vector Search: User queries are matched against embeddings using FAISS
4. 4.LLM Synthesis: Groq/Ollama generates coherent answers from retrieved context
5. 5.Source Attribution: Answers include references to source files

🎯 LLM Provider Options

Groq (Recommended) ⚡

• —Speed: 300+ tokens/second
• —Cost: Free tier available
• —Quality: Llama 3.3 70B
• —Setup: Get API key from https://console.groq.com

Ollama (Local Alternative) 🏠

• —Speed: 20-30 tokens/second
• —Cost: 100% free
• —Quality: Llama 3.2 8B
• —Setup: Install Ollama locally

📝 Customization

Update Content

Simply add/edit markdown files in data/markdown_files/ and restart the application. The index will be rebuilt automatically.

Adjust Chunk Size

Edit src/data_loader.py:

def _create_chunks(self, chunk_size: int = 800, overlap: int = 150)

Change System Prompt

Edit src/llm_generator.py in the _get_system_prompt() method to customize the chatbot's personality and behavior.

Customize UI

Edit src/main.py to modify the Gradio interface, colors, examples, and branding.

🚀 Deployment Options

Hugging Face Spaces

1. 1.Create account at https://huggingface.co
2. 2.Create new Space (Gradio)
3. 3.Upload project files
4. 4.Add GROQ_API_KEY to Space secrets

Docker

docker build -t website-chatbot .
docker run -p 7860:7860 --env-file .env website-chatbot

Cloud Platforms

• —Render: Deploy as web service
• —Railway: One-click deploy
• —AWS/GCP/Azure: Deploy as container

🛠️ Development

Test Individual Components

# Test data loader
python src/data_loader.py

# Test embedding indexer
python src/embedding_indexer.py

# Test search engine
python src/search.py

# Test LLM generator
python src/llm_generator.py

📊 Performance

• —Index Creation: ~30 seconds for 19 markdown files
• —Query Response: 1-3 seconds with Groq
• —Memory Usage: ~500MB with loaded index
• —Concurrent Users: Supports multiple simultaneous queries

🔒 Security Notes

• —Never commit .env file with real API keys
• —Use environment variables for sensitive data
• —Consider rate limiting for production deployment
• —Validate and sanitize user inputs

📄 License

MIT License - Feel free to use for commercial projects

🤝 Support

For issues or questions:

1. 1.Check the markdown files are properly formatted
2. 2.Verify API key is set correctly in .env
3. 3.Ensure all dependencies are installed
4. 4.Check logs for error messages

🎓 Based On

This project is adapted from the Florida Masonic Law Assistant, demonstrating the flexibility of RAG systems for different content domains.