📚 Document Q&A System

A powerful document question-answering system built with LlamaIndex and Gradio. Upload your documents and ask questions about them using state-of-the-art AI models.

Features

🔍 Smart Document Processing: Automatically processes various document formats (PDF, TXT, DOCX, MD, CSV, JSON)

🤖 Multiple AI Models: Choose from GPT-4o, Claude 3.5 Sonnet, Llama 3.1, Mistral, and more

📊 Performance Monitoring: Track response times and query statistics

🎯 Source Attribution: See which document sections were used to generate answers

⚙️ Customizable Settings: Adjust temperature, token limits, and retrieval parameters

🔒 Secure API Key Management: Use environment variables or direct input

How to Use

1. Upload Documents

Go to the "Upload Documents" tab
Select your files (PDF, TXT, DOCX, MD, CSV, JSON)
Click "Process Documents" to create the searchable index

2. Configure Settings

Add your OpenRouter API key (or set as HF Space secret)
Choose your preferred AI model
Adjust parameters like temperature and max tokens

3. Ask Questions

Enter your question in the "Ask Questions" tab
Click "Ask Question" to get AI-powered answers
View sources and performance metrics

API Key Setup

You can provide your OpenRouter API key in two ways:

Direct Input: Enter it in the "API Key" field in the interface
Environment Variable: Set OPENROUTER_API_KEY as a Hugging Face Space secret

Get your API key from OpenRouter

Best Practices for Questions

🎯 Be specific: "What does the author say about climate change?" vs "Tell me about climate"
📚 Ask about concepts: "What is the main methodology discussed?"
🔍 Use comparative questions: "How do different studies approach this topic?"
📊 Request analysis: "What are the key findings presented?"
🏛️ Ask about methodology: "What research methods are used?"

Available Models

GPT-4o: Best overall performance, most accurate
GPT-4o Mini: Faster, cost-effective option
Claude 3.5 Sonnet: Excellent reasoning and analysis
Claude 3 Haiku: Fast and efficient
Llama 3.1 70B/8B: Open source, strong performance
Mistral Large: Strong multilingual capabilities
Gemini Pro: Google's advanced model

Technical Details

Built with:

LlamaIndex: Document indexing and retrieval
Gradio: Web interface
OpenRouter: Multi-model API access
HuggingFace Embeddings: Text vectorization
BGE-small-en-v1.5: Efficient embedding model

Performance

Vector-based semantic search for accurate retrieval
Cached indexing for fast subsequent queries
Configurable chunk sizes and overlap for optimal results
Real-time performance monitoring

Development

To run locally:

git clone <your-repo>
cd document-qa-system
pip install -r requirements.txt
python app.py

License

This project is open source and available under the MIT License.

Support

For issues or questions, please check the Help tab in the application or create an issue in the repository.