📚 RAG Applications with NVIDIA NeMo and Streamlit

🌟 Overview

A sophisticated implementation of Retrieval Augmented Generation (RAG) leveraging NVIDIA's AI endpoints and Streamlit. This application transforms document analysis and question-answering through state-of-the-art language models and efficient vector search capabilities.

🎯 Core Features

• 📄 Multi-PDF Document Processing
• 🔍 Advanced Text Chunking System
• 💾 FAISS Vector Store Integration
• ⚡ NVIDIA NIM Endpoints
• 🤖 Llama 3.1 405B Model Support
• ⏱️ Real-time Performance Metrics
• 📊 Similarity Search Visualization

🛠️ Technical Architecture

Component Stack

graph TD
    A[PDF Documents] --> B[Document Processor]
    B --> C[Text Chunker]
    C --> D[NVIDIA Embeddings]
    D --> E[FAISS Vector Store]
    F[User Query] --> G[Query Processor]
    G --> E
    E --> H[NVIDIA LLM]
    H --> I[Response Generator]

Loading

System Requirements

• Python 3.8+
• 8GB RAM minimum
• NVIDIA API access
• Internet connectivity
• PDF processing capabilities

📦 Installation & Setup

1. Clone Repository

git clone https://github.com/arsath-eng/RAG1-NVIDIA-GENAI.git
cd RAG1-NVIDIA-GENAI

2. Environment Setup

# Create virtual environment
python -m venv venv

# Activate environment
# For Windows
.\venv\Scripts\activate
# For Unix/Mac
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

3. Configuration

# Create environment file
touch .env

# Add required credentials
NVIDIA_API_KEY=your_api_key_here

💻 Usage Guide

Application Launch

streamlit run app.py

Document Processing Workflow

1. Document Upload

   • Support for multiple PDF files
   • Automatic text extraction
   • Progress tracking

2. Embedding Creation

   • Click "Create Document Embeddings"
   • Automatic chunking and processing
   • Vector store initialization

3. Query Processing

   • Enter questions about documents
   • Real-time response generation
   • View similarity search results

   

## ⚙️ Configuration Options

Text Splitting Parameters

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=700,  # Adjust for document length
    chunk_overlap=50  # Modify for context preservation
)

Model Configuration

llm = ChatNVIDIA(
    model="meta/llama-3.1-405b-instruct",
    temperature=0.7,  # Adjust for response creativity
    max_tokens=512    # Modify for response length
)

📈 Performance Optimization

Vector Store Tuning

• Optimal chunk size selection
• Embedding dimension management
• Index optimization techniques

Response Time Improvements

• Query preprocessing
• Cache implementation
• Batch processing capabilities

🔍 Advanced Features

1. Intelligent Chunking

• Content-aware text splitting
• Semantic boundary preservation
• Overlap optimization

2. Vector Search Enhancement

• Nearest neighbor search
• Similarity threshold tuning
• Result ranking optimization

3. Response Generation

• Context-aware answers
• Source attribution
• Confidence scoring

🚀 Best Practices

Document Preparation

• Use clear, well-formatted PDFs
• Ensure text is extractable
• Optimize document size

Query Formulation

• Be specific and clear
• Include relevant context
• Use natural language

System Optimization

• Monitor memory usage
• Regular cache clearing
• Performance tracking

🛠️ Troubleshooting

Common Issues

1. PDF Processing Errors

   • Solution: Check PDF format compatibility
   • Verify text extraction capabilities

2. Memory Issues

   • Solution: Adjust chunk size
   • Implement batch processing

3. API Connection

   • Solution: Verify credentials
   • Check network connectivity

🤝 Contributing

Development Workflow

1. Fork repository
2. Create feature branch

git checkout -b feature/YourFeature

3. Commit changes

git commit -m 'Add YourFeature'

4. Push to branch

git push origin feature/YourFeature

5. Submit Pull Request

📝 License

This project is licensed under the MIT License. See LICENSE for details.

🙏 Acknowledgments

• NVIDIA AI Team
• Streamlit Community
• LangChain Contributors
• Meta AI Research

📞 Support & Contact

• Create an Issue
• Join Discussions
• Review Documentation

---

Made with ❤️ by @arsath-eng