📚 LangChain Document Q&A System

An intelligent containerized document question-answering system powered by LangChain

Overview

Transform your documents into an interactive knowledge base! This application allows you to upload PDF or text documents and ask intelligent questions about their content using state-of-the-art AI models. All in a docker container!

Architecture

graph TD
    A[Document Upload] --> B[Document Processor]
    B --> C[Text Splitting]
    C --> D[Bedrock Embeddings]
    D --> E[ChromaDB Vector Store]
    E --> F[Retrieval QA System]
    F --> G[Claude Sonnet Response]
    G --> H[Streamlit UI]

Loading

📦 Installation

1. Build Docker Image

   docker build -t pdf-chat .

2. Run Docker Container

   docker compose up

3. Open your browser Navigate to http://localhost:8501

Configuration

Model Configuration

The system uses:

• LLM: us.anthropic.claude-sonnet-4-20250514-v1:0
• Embeddings: amazon.titan-embed-text-v1
• Vector Store: ChromaDB with persistent storage

Usage

1. Upload Document 📄

• Click "Upload a document"
• Select your PDF or TXT file
• Wait for processing to complete

2. Ask Questions 💭

• Type your question in the text input
• Get comprehensive answers (250+ words)
• View source context and references

Project Structure

pdf-chat/
├── main.py                 # Streamlit application entry point
├── qa_system.py           # Q&A system implementation
├── document_loader.py     # Document processing utilities
├── requirements.txt       # Python dependencies
├── refdocs/              # Uploaded documents storage
├── chroma_db/            # Vector database storage

Technical Details

Document Processing Pipeline

1. Loading: PyPDFLoader for PDFs, TextLoader for text files
2. Chunking: RecursiveCharacterTextSplitter
3. Embedding: AWS Bedrock Titan embeddings
4. Storage: ChromaDB vector database with persistence

Q&A System

• Retrieval: Similarity search with top-k=3 results
• Generation: Claude Sonnet with custom prompt template
• Output: Detailed 250+ word responses with context

Customization

Modify Chunk Size

# In document_loader.py
self.text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=1500,  # Increase for larger chunks
    chunk_overlap=300
)

Adjust Model Parameters

# In qa_system.py
self.llm = ChatBedrock(
    model="us.anthropic.claude-sonnet-4-20250514-v1:0",
    model_kwargs={
        "max_tokens": 2000,    # Increase for longer responses
        "temperature": 0.3,    # Adjust creativity (0-1)
        "top_p": 0.9
    }
)

License

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

Acknowledgments

• LangChain - For the amazing framework
• AWS Bedrock - For powerful AI models
• Streamlit - For the UI framework
• ChromaDB - For efficient vector storage
• Docker - For containerization