Document Intelligence Assistant

A powerful Streamlit application that transforms your documents into an intelligent chatbot using Docling for document processing and LangGraph for conversational AI.

Overview

This application allows you to upload various document formats (PDF, Word, PowerPoint, HTML) and interact with them through a conversational AI interface. It combines IBM's Docling library for advanced document understanding with LangGraph for building sophisticated AI agents.

Features

📄 Multi-Format Document Support

• PDF documents - Full text extraction with OCR support
• Word documents (.docx) - Complete text and structure preservation
• PowerPoint presentations (.pptx) - Slide content extraction
• HTML files - Web content processing

💬 Intelligent Chat Interface

Ask questions about your documents and get accurate, context-aware answers powered by OpenAI's language models.

📊 Document Structure Visualization

Explore your documents in detail with multiple visualization tabs:

• Summary - Overview of pages, tables, images, and content types
• Hierarchy - Document outline and heading structure
• Tables - Interactive table extraction and display
• Images - Picture extraction with captions and positioning

🔍 Advanced Document Processing

• Structure preservation - Maintains document hierarchy, tables, and formatting
• OCR support - Handles scanned documents automatically
• Smart chunking - Optimized text segmentation for better retrieval
• Vector search - Fast and accurate information retrieval using ChromaDB

🤖 LangGraph AI Agent

• Contextual understanding - Maintains conversation history
• Tool integration - Uses vector search to find relevant information
• Streaming responses - Real-time answer generation

Installation

Prerequisites

• Python 3.8 or higher
• OpenAI API key

Setup

1. Clone the repository:

git clone <repository-url>
cd app

2. Install dependencies:

pip install -r requirements.txt

3. Configure environment variables:

cp .env.example .env

Edit .env and add your OpenAI API key:

OPENAI_API_KEY=your_api_key_here

Usage

1. Start the Streamlit application:

streamlit run app.py

2. Upload your documents:

   • Click on the file uploader in the sidebar
   • Select one or more documents (PDF, DOCX, PPTX, or HTML)
   • Click "Process & Index" to process the documents

3. Start chatting:

   • Once processing is complete, navigate to the "Chat" tab
   • Ask questions about your documents
   • Get intelligent, context-aware answers

4. Explore document structure:

   • Switch to the "Document Structure" tab
   • Select a document to analyze
   • View detailed information about content, hierarchy, tables, and images

Project Structure

app/
├── app.py                      # Main Streamlit application
├── requirements.txt            # Python dependencies
├── .env.example               # Environment variables template
├── .gitignore                # Git ignore file
└── src/
    ├── __init__.py           # Package initializer
    ├── agent.py              # LangGraph agent configuration
    ├── document_processor.py # Docling document processing
    ├── structure_visualizer.py # Document structure analysis
    ├── tools.py              # Agent tools definition
    └── vectorstore.py        # Vector store management

Key Dependencies

• docling (>=2.55.0) - Document processing and structure extraction
• langchain (>=0.3.0) - LLM framework and utilities
• langchain-docling (>=0.1.0) - Docling integration for LangChain
• langchain-openai (>=0.2.0) - OpenAI integration
• langgraph (>=0.2.0) - Agent orchestration and workflow
• langchain-chroma (>=0.1.0) - ChromaDB vector store integration
• streamlit (>=1.28.0) - Web application framework
• chromadb (>=0.4.22) - Vector database for embeddings
• pandas (>=2.0.0) - Data manipulation and display

How It Works

1. Document Upload: Users upload documents through the Streamlit interface

2. Processing with Docling: Documents are processed using Docling, which:

   • Extracts text with OCR support
   • Identifies document structure (headings, paragraphs, tables)
   • Preserves layout and formatting
   • Extracts images and their captions

3. Text Chunking: Processed documents are split into optimized chunks for better retrieval

4. Vector Store Creation: Text chunks are embedded and stored in ChromaDB for semantic search

5. Agent Creation: A LangGraph agent is configured with:

   • Access to the vector store through a search tool
   • Conversation memory for context
   • Streaming capabilities for real-time responses

6. Conversation: Users interact with the agent, which:

   • Searches the vector store for relevant information
   • Generates contextual answers based on document content
   • Maintains conversation history for follow-up questions

Production Considerations

For production deployment, consider:

• Persistent storage - Use a persistent vector database instead of in-memory storage
• Batch processing - Handle large document collections more efficiently
• GPU acceleration - Speed up OCR and document processing
• Authentication - Add user authentication and access controls
• Caching - Implement caching for processed documents
• Rate limiting - Add API rate limiting for OpenAI calls
• Error handling - Enhanced error recovery and logging
• Monitoring - Add application monitoring and analytics

License

This project is provided as-is for educational and demonstration purposes.

Acknowledgments

• Docling by IBM Research for advanced document understanding
• LangChain for the LLM framework
• LangGraph for agent orchestration
• Streamlit for the web interface
• OpenAI for language models