CrewAI Agentic RAG System

An intelligent Retrieval-Augmented Generation (RAG) system built with CrewAI, featuring custom PDF document search, web search integration, and multiple user interfaces. This project demonstrates advanced agent orchestration for information retrieval and synthesis.

🌟 Features

Core Functionality

• Multi-Source Information Retrieval: Combines PDF document search with web search capabilities
• Agent-Based Architecture: Uses specialized agents for retrieval and response synthesis
• Vector Database Integration: Employs Qdrant for efficient document chunk storage and similarity search
• Flexible Search Strategy: Prioritizes PDF content, falls back to web search when needed
• Sequential Processing: Ensures coherent information flow between agents

User Interfaces

• Command-Line Interface: Simple CLI for quick queries (main.py)
• Web Interface: Interactive Streamlit app with PDF upload and chat functionality (app_llama.py)
• Real-time Streaming: Progressive response display in web interface

Technical Features

• Custom PDF Processing: Uses MarkItDown for robust PDF text extraction
• Intelligent Chunking: Recursive text splitting with overlap for optimal retrieval
• Multiple LLM Support: Configurable language models (Ollama, OpenAI, etc.)
• Tool Integration: SerperDev and Firecrawl for web search
• Environment Configuration: Secure API key management with dotenv

🏗️ Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   User Query    │───▶│  Retriever Agent │───▶│ Response Synth  │
│                 │    │                  │    │   Agent         │
└─────────────────┘    └──────────────────┘    └─────────────────┘
                              │                           │
                              ▼                           ▼
                       ┌─────────────┐            ┌─────────────┐
                       │ PDF Search  │            │   Synthesis │
                       │   Tool      │            │             │
                       └─────────────┘            └─────────────┘
                              │
                              ▼
                       ┌─────────────┐
                       │ Web Search  │
                       │   Tool      │
                       └─────────────┘

Components

Agents

• Retriever Agent: Handles information gathering from PDF and web sources
• Response Synthesizer Agent: Processes retrieved information into coherent responses

Tools

• DocumentSearchTool: Custom tool for PDF content search using Qdrant
• SerperDevTool: Web search integration
• FirecrawlSearchTool: Advanced web scraping and search

Configuration

• agents.yaml: Agent role definitions and backstories
• tasks.yaml: Task descriptions and expected outputs

🚀 Installation

Prerequisites

• Python 3.8 or higher
• Ollama (for local LLM support)
• Git

Setup Steps

1. Clone the repository

   git clone https://github.com/Monish-Nallagondalla/crewai_agentic_rag.git
   cd crewai_agentic_rag
   

2. Create virtual environment

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

3. Install dependencies

   pip install -r requirements.txt
   

4. Set up environment variables

   cp .env.example .env  # Create if not exists
   

   Edit .env with your API keys:

   SERPER_API_KEY=your_serper_api_key
   FIRECRAWL_API_KEY=your_firecrawl_api_key
   

5. Start Ollama (for local LLM)

   ollama serve
   ollama pull llama3.2  # Pull the required model
   

📖 Usage

Command-Line Interface

Run a simple query:

python src/agentic_rag/main.py

The default query is: "Who is Elon Musk and what is his net worth?"

Web Interface

1. Start the Streamlit app

   streamlit run app_llama.py
   

2. Access the interface

   • Open http://localhost:8501 in your browser
   • Upload a PDF document in the sidebar
   • Wait for indexing to complete
   • Start chatting with your document

Features in Web Interface

• PDF Upload: Drag and drop PDF files for indexing
• Real-time Chat: Interactive conversation with the RAG system
• Progress Indicators: Visual feedback during processing
• Chat History: Persistent conversation history

⚙️ Configuration

Agent Configuration (src/agentic_rag/config/agents.yaml)

retriever_agent:
  role: "Retrieve relevant information..."
  goal: "Retrieve the most relevant information..."
  backstory: "You're a meticulous analyst..."

response_synthesizer_agent:
  role: "Response synthesizer agent..."
  goal: "Synthesize the retrieved information..."
  backstory: "You're a skilled communicator..."

Task Configuration (src/agentic_rag/config/tasks.yaml)

retrieval_task:
  description: "Retrieve the most relevant information..."
  expected_output: "The most relevant information in form of text..."
  agent: retriever_agent

response_task:
  description: "Synthesize the final response..."
  expected_output: "A concise and coherent response..."
  agent: response_synthesizer_agent

Customizing the PDF Search Tool

Modify chunk size and overlap in src/agentic_rag/tools/custom_tool.py:

text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=512,      # Adjust chunk size
    chunk_overlap=50,    # Adjust overlap
    # ... other parameters
)

🔧 Development

Project Structure

crewai_agentic_rag/
├── src/
│   └── agentic_rag/
│       ├── __init__.py
│       ├── crew.py              # Main CrewAI setup
│       ├── main.py              # CLI entry point
│       ├── config/
│       │   ├── agents.yaml      # Agent configurations
│       │   └── tasks.yaml       # Task configurations
│       └── tools/
│           ├── __init__.py
│           └── custom_tool.py   # Custom PDF search tool
├── app_llama.py                 # Streamlit web interface
├── requirements.txt             # Python dependencies
├── .gitignore                   # Git ignore rules
├── LICENSE                      # MIT License
├── README.md                    # This file
├── assets/                      # Static assets
│   └── deep-seek.png
├── knowledge/                   # Knowledge base documents
│   └── dspy.pdf
└── flow_diagram.svg             # Architecture diagram

Adding New Tools

1. Create a new tool class in src/agentic_rag/tools/
2. Inherit from BaseTool and implement the required methods
3. Add the tool to the agent's tool list in crew.py

Extending Agents

1. Add new agent configuration in agents.yaml
2. Create the agent method in crew.py with @agent decorator
3. Define corresponding tasks in tasks.yaml

🤝 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

Development Guidelines

• Follow PEP 8 style guidelines
• Add type hints for new functions
• Write comprehensive docstrings
• Add unit tests for new features
• Update documentation for API changes

📊 Performance & Metrics

Current Capabilities

• PDF Processing: Handles various PDF formats using MarkItDown
• Chunking Strategy: 512-character chunks with 50-character overlap
• Vector Storage: In-memory Qdrant for fast retrieval
• Search Accuracy: Semantic similarity-based retrieval
• Response Quality: Agent-based synthesis for coherent answers

Known Limitations

• In-memory vector storage (suitable for small documents)
• Single PDF support per session (web interface)
• Sequential processing (no parallel retrieval)

🔒 Security

• API keys stored securely in environment variables
• No sensitive data logged in application output
• Input validation for file uploads
• Secure temporary file handling

📝 License

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

🙏 Acknowledgments

• CrewAI - Multi-agent framework
• Qdrant - Vector database
• Streamlit - Web app framework
• Ollama - Local LLM runtime
• MarkItDown - PDF processing

📞 Support

For questions, issues, or contributions:

• Open an issue on GitHub
• Check the documentation in this README
• Review the code comments for implementation details

Made with ❤️ using CrewAI and modern AI tools