🧠 DocuMind AI

DocuMind AI is an intelligent multi-agent document analysis system that leverages AWS Bedrock's powerful AI models to provide comprehensive document insights through a collaborative agent architecture.

🌟 Features

• Multi-Agent Architecture: Specialized agents for retrieval, analysis, and validation
• PDF Document Processing: Upload and analyze multiple PDF documents simultaneously
• Intelligent Retrieval: Vector-based similarity search using FAISS
• Advanced Analysis: Powered by Claude 3 Sonnet for deep document understanding
• Response Validation: Cross-validation using Llama 3 70B for accuracy assurance
• Real-time Processing: Live progress tracking and performance metrics
• Intuitive Interface: Beautiful Streamlit-based UI with responsive design

🏗️ Architecture Overview

DocuMind AI implements a sophisticated multi-agent system where each agent specializes in a specific task:

graph TB
    A[User Upload PDFs] --> B[Document Processor]
    B --> C[Text Splitting & Chunking]
    C --> D[Vector Embeddings]
    D --> E[FAISS Vector Store]
    
    F[User Query] --> G[Multi-Agent Orchestrator]
    G --> H[Retrieval Agent]
    G --> I[Analysis Agent]
    G --> J[Validation Agent]
    
    H --> E
    E --> H
    H --> K[Retrieved Documents]
    
    K --> I
    I --> L[Claude 3 Analysis]
    L --> M[Analysis Response]
    
    M --> J
    K --> J
    J --> N[Llama 3 Validation]
    N --> O[Final Validated Response]
    
    O --> P[User Interface]
    
    style G fill:#e1f5fe
    style H fill:#f3e5f5
    style I fill:#e8f5e8
    style J fill:#fff3e0

Loading

🔧 System Components

1. Document Processor

• Purpose: Handles PDF upload and preprocessing
• Technology: PyPDFLoader for extraction
• Features:
  • Multi-file processing
  • Metadata extraction
  • Error handling and validation

2. Retrieval Agent 🔍

• Model: Amazon Titan Text Embeddings v2
• Purpose: Finds relevant document chunks
• Process:
  • Converts queries to vector embeddings
  • Performs similarity search in FAISS
  • Returns top-k relevant documents with scores

3. Analysis Agent 🧠

• Model: Claude 3 Sonnet (anthropic.claude-3-sonnet-20240229-v1:0)
• Purpose: Provides comprehensive document analysis
• Capabilities:
  • Deep content understanding
  • Contextual question answering
  • Insight extraction and summarization

4. Validation Agent ✅

• Model: Llama 3 70B Instruct (meta.llama3-70b-instruct-v1:0)
• Purpose: Validates analysis accuracy
• Features:
  • Cross-reference with source documents
  • Accuracy scoring (1-10 scale)
  • Confidence level assessment

5. Multi-Agent Orchestrator 🎭

• Purpose: Coordinates the entire workflow
• Process Flow:
  1. Route query to Retrieval Agent
  2. Pass results to Analysis Agent
  3. Send analysis for validation
  4. Aggregate results and metrics

🏛️ Technical Architecture

graph LR
    subgraph "Frontend Layer"
        A[Streamlit UI]
        B[File Upload]
        C[Query Interface]
        D[Results Display]
    end
    
    subgraph "Application Layer"
        E[Multi-Agent Orchestrator]
        F[Document Processor]
        G[Session Management]
    end
    
    subgraph "Agent Layer"
        H[Retrieval Agent]
        I[Analysis Agent]
        J[Validation Agent]
    end
    
    subgraph "Storage Layer"
        K[FAISS Vector Store]
        L[Temporary File Storage]
    end
    
    subgraph "AWS Bedrock"
        M[Titan Embeddings]
        N[Claude 3 Sonnet]
        O[Llama 3 70B]
    end
    
    A --> E
    B --> F
    C --> E
    E --> H
    E --> I
    E --> J
    F --> K
    H --> M
    I --> N
    J --> O
    H --> K
    
    style E fill:#ffeb3b
    style H fill:#2196f3
    style I fill:#4caf50
    style J fill:#ff9800

Loading

🚀 Getting Started

Prerequisites

• Python 3.8 or higher
• AWS Account with Bedrock access
• AWS CLI configured with appropriate credentials

AWS Bedrock Model Access

Ensure you have access to the following models in AWS Bedrock:

• amazon.titan-embed-text-v2:0
• anthropic.claude-3-sonnet-20240229-v1:0
• meta.llama3-70b-instruct-v1:0

Installation

1. Clone the repository

   git clone https://github.com/yourusername/documind-ai.git
   cd documind-ai

2. Create virtual environment

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

3. Install dependencies

   pip install -r requirements.txt

4. Configure AWS credentials

   aws configure
   # Enter your AWS Access Key ID, Secret Access Key, and region (us-east-1)

Running the Application

Local Development

streamlit run app.py

The application will be available at http://localhost:8501

Production Deployment

The app is deployed on Streamlit Cloud and accessible at: DocuMind AI · Streamlit

📋 Usage Guide

Step 1: Document Upload

1. Use the sidebar file uploader
2. Select one or more PDF files
3. Click "🚀 Process Documents"
4. Wait for processing completion

Step 2: Query Documents

1. Enter your question in the main input field
2. Click "🔍 Analyze"
3. View the multi-agent processing progress
4. Review the comprehensive analysis

Step 3: Explore Results

• Main Analysis: Primary response from Claude 3
• Validation: Quality check from Llama 3
• Processing Details: Performance metrics and statistics

🔧 Configuration

Environment Variables

AWS_REGION=us-east-1
EMBED_MODEL_ID=amazon.titan-embed-text-v2:0
CLAUDE_MODEL_ID=anthropic.claude-3-sonnet-20240229-v1:0
LLAMA_MODEL_ID=meta.llama3-70b-instruct-v1:0

Model Parameters

• Embedding Dimensions: 1024 (Titan v2)
• Text Chunk Size: 1000 characters
• Chunk Overlap: 200 characters
• Max Tokens: 1000 (Analysis), 500 (Validation)

📊 Performance Metrics

The system tracks various performance indicators:

• Processing Time: End-to-end query processing
• Document Retrieval: Number of relevant chunks found
• Relevance Scores: Similarity matching accuracy
• Validation Scores: Response quality assessment (1-10)

🛠️ Troubleshooting

Common Issues

1. AWS Credentials Error

   Solution: Ensure AWS CLI is configured with valid credentials
   aws configure list
   

2. Model Access Denied

   Solution: Request access to Bedrock models in AWS Console
   Navigate to: AWS Bedrock → Model Access → Request Access
   

3. PDF Processing Error

   Solution: Ensure PDFs are not password-protected and under 10MB
   

4. FAISS Index Error

   Solution: Re-upload documents to rebuild the vector index
   

🧪 Testing

Sample Test Queries

• "What are the main themes discussed in the document?"
• "Summarize the key findings and recommendations"
• "What are the limitations mentioned in the study?"
• "Extract all numerical data and statistics"

🤝 Contributing

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

📝 License

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

🙏 Acknowledgments

• AWS Bedrock for providing state-of-the-art AI models
• Streamlit for the excellent web framework
• LangChain for the AI application framework
• FAISS for efficient vector similarity search

👨‍💻 Author

Jay Zalani

• GitHub: @jayzalani
• LinkedIn: Jay Zalani

---

Made with ❤️ by Jay Zalani

🚀 Try DocuMind AI Live