🔍 DeepSeek RAG - Local Document Chat

A Retrieval-Augmented Generation (RAG) system that allows you to chat with your local documents using DeepSeek's powerful language model. This project demonstrates how to build an intelligent document Q&A system using modern AI technologies.

🚀 Features

• Local Document Processing: Automatically processes MDX and other document formats
• Vector Database: Uses ChromaDB for efficient document embeddings and retrieval
• DeepSeek Integration: Leverages DeepSeek-Coder model for intelligent responses
• Streamlit Interface: Clean and intuitive web interface for document interaction
• Smart Text Processing: Cleans MDX files and splits documents into optimal chunks

🛠️ Tech Stack

• Language Model: DeepSeek-Coder (via Ollama)
• Vector Database: ChromaDB
• Embeddings: HuggingFace sentence-transformers/all-MiniLM-L6-v2
• Framework: LangChain
• Frontend: Streamlit
• Document Processing: UnstructuredFileLoader, CharacterTextSplitter

📋 Prerequisites

Before running this project, ensure you have:

1. Python 3.8+ installed
2. Ollama installed and running
3. DeepSeek-Coder model pulled in Ollama

Installing Ollama and DeepSeek-Coder

# Install Ollama (macOS)
brew install ollama

# Or download from https://ollama.ai

# Pull the DeepSeek-Coder model
ollama pull deepseek-coder

🚀 Installation

1. Clone the repository

   git clone <your-repo-url>
   cd deepseek-rag

2. Create a virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

📖 Usage

1. Prepare your documents

   • Place your documents in the docs/ folder
   • Supports MDX files and other formats (PDF, TXT, etc.)
   • The system will automatically process and clean MDX content

2. Start the application

   streamlit run app.py

3. Open your browser

   • Navigate to http://localhost:8501
   • Start asking questions about your documents!

🔧 Configuration

Customizing the RAG Chain

You can modify the RAG configuration in rag_chain.py:

# Adjust chunk size and overlap
splitter = CharacterTextSplitter(chunk_size=500, chunk_overlap=50)

# Change embedding model
embedding = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

# Switch to different DeepSeek model
llm = Ollama(model="deepseek-coder")  # or "deepseek-chat"

Document Processing

The system automatically:

• Cleans HTML tags from MDX files
• Splits documents into chunks for better retrieval
• Creates vector embeddings for semantic search
• Persists the vector database in the db/ directory

📁 Project Structure

deepseek-rag/
├── app.py                 # Streamlit web interface
├── rag_chain.py          # RAG implementation and configuration
├── requirements.txt      # Python dependencies
├── docs/                 # Document directory (Next.js docs included)
│   ├── 01-installation.mdx
│   ├── 02-project-structure.mdx
│   └── ...
└── db/                   # ChromaDB vector database
    ├── chroma.sqlite3
    └── ...

🔍 How It Works

1. Document Loading: The system scans the docs/ folder and loads all supported documents
2. Text Processing: MDX files are cleaned of HTML tags, and all documents are split into chunks
3. Embedding Creation: Text chunks are converted to vector embeddings using HuggingFace models
4. Vector Storage: Embeddings are stored in ChromaDB for fast similarity search
5. Query Processing: User questions are embedded and matched against document chunks
6. Response Generation: DeepSeek-Coder generates answers based on retrieved context

🎯 Example Use Cases

This RAG system is perfect for:

• Documentation Q&A: Ask questions about technical documentation
• Knowledge Base: Build internal company knowledge systems
• Research Assistant: Query research papers and technical documents
• Learning Tool: Interactive way to explore complex documentation

🤝 Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

📝 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• DeepSeek for the powerful language models
• LangChain for the RAG framework
• ChromaDB for vector database capabilities
• Streamlit for the user interface
• HuggingFace for embedding models

🐛 Troubleshooting

Ollama Connection Issues:

• Ensure Ollama is running: ollama serve
• Verify DeepSeek model is available: ollama list

Memory Issues:

• Reduce chunk_size in rag_chain.py
• Use a smaller embedding model

Document Processing Errors:

• Check document formats are supported
• Ensure documents are readable and not corrupted

---

Built with ❤️ using DeepSeek, LangChain, and Streamlit