Agentic RAG Application with CrewAI and Streamlit

This project demonstrates an agentic Retrieval-Augmented Generation (RAG) application built using CrewAI and Streamlit. It allows users to upload a PDF document and then ask questions about its content, leveraging the power of Large Language Models (LLMs) and intelligent agents.

Project Preview

Features

• PDF Document Upload: Users can upload PDF files directly through the Streamlit interface.
• PDF Indexing: The application indexes the uploaded PDF, making its content searchable.
• Agentic RAG: Employs a CrewAI-powered crew of agents to handle information retrieval and response generation.
• Web Search Fallback: If information is not found in the PDF, the system intelligently falls back to web search using the Serper API.
• Conversational Interface: Provides a user-friendly chat interface for interacting with the system.
• Clear Chat History: Users can easily clear the chat history and start fresh.
• Real-time Response: The application displays the LLM's response in a streaming fashion, providing a more engaging user experience.
• Local LLM Support: Uses Ollama to run a local LLM (deepseek-r1:14b)

How It Works

The application uses a crew of two agents:

1. Retriever Agent:
   • Responsible for retrieving relevant information based on the user's query.
   • Prioritizes searching the uploaded PDF.
   • Falls back to web search if the PDF doesn't contain the answer.
   • Uses PDFSearchTool and SerperDevTool.
2. Response Agent:
   • Synthesizes the retrieved information into a clear and concise response.
   • Handles cases where no relevant information is found.

These agents work together sequentially to provide accurate and helpful answers to user queries.

Technologies Used

• CrewAI: A framework for orchestrating multi-agent systems.
• Streamlit: A Python library for creating interactive web applications.
• Ollama: For running the local LLM.
• Deepseek-r1:14b: The specific LLM used in this project.
• Serper API: For web search capabilities.
• PDFSearchTool: A custom tool for searching PDF documents.
• Python: The primary programming language.
• dotenv: For managing environment variables.

Prerequisites

• Python 3.8+
• Ollama: You need to have Ollama installed and running locally.
• Serper API Key: You'll need a Serper API key for web search functionality.
• Required Python Packages:

  pip install streamlit crewai crewai-tools python-dotenv

Installation and Setup

1. Clone the Repository:

   git clone <your-repository-url>
   cd <your-repository-directory>

2. Install Dependencies:

   pip install -r requirements.txt

   (You can create a requirements.txt file with the following content:

   streamlit
   crewai
   crewai-tools
   python-dotenv
   

   )
3. Set Environment Variables:
   • Create a .env file in the project's root directory.
   • Add your Serper API key:

     SERPER_API_KEY=your_serper_api_key_here
     

4. Run Ollama:
   • Make sure you have Ollama installed and running.
   • Pull the deepseek-r1:14b model:

   ollama pull deepseek-r1:14b

5. Run the Streamlit App:

   streamlit run rag_app.py

Usage

1. Upload a PDF: In the sidebar, use the file uploader to select and upload a PDF document.
2. Wait for Indexing: The application will index the PDF. This may take a few moments.
3. Ask Questions: In the chat interface, type your questions about the PDF content.
4. View Responses: The system will provide answers based on the PDF content or, if necessary, from the web.
5. Clear Chat: Use the "Clear Chat" button in the sidebar to reset the conversation.

---