Agentic RAG Chatbot

Live Link: https://huggingface.co/spaces/philip11/Agentic-RAG-Chatbot

Developed by: PHILIP SIMON DEROCK P

This project implements a sophisticated, multi-agent Retrieval-Augmented Generation (RAG) chatbot. It allows users to upload various document types and ask questions about their content. The chatbot leverages a powerful agentic architecture, where specialized agents collaborate to ingest, retrieve, and generate responses, all orchestrated through a central message bus.

✨ Features

• Multi-Format Document Support: Upload and chat with PDFs, PowerPoints, CSVs, Word documents, Markdown, and plain text files.
• Retrieval-Augmented Generation (RAG): Answers are generated based on the content of your uploaded documents, ensuring relevance and accuracy.
• Agentic Architecture: A modular system where different agents handle specific tasks (ingestion, retrieval, LLM response), making the system scalable and maintainable.
• Vector-Based Semantic Search: Utilizes ChromaDB and Sentence-Transformers for efficient and accurate semantic search across your documents.
• Google Gemini Integration: Powered by Google's gemini-1.5-flash-latest model for high-quality response generation.
• Interactive UI: A user-friendly web interface built with Streamlit for easy document management and conversation.

🏗️ Architecture Overview

The chatbot operates on a Model Context Protocol (MCP), an internal message bus that facilitates communication between specialized agents.

1. UI Agent (UIAgent): Manages the Streamlit front end, handling user interactions like file uploads and queries.
2. Ingestion Agent (IngestionAgent): Receives documents from the UI, parses them using DocumentParser, and sends the extracted text chunks to the Retrieval Agent.
3. Retrieval Agent (RetrievalAgent): Manages the VectorStore. It receives text chunks, creates vector embeddings, and stores them in ChromaDB. It also handles search queries to find relevant context.
4. LLM Response Agent (LLMResponseAgent): Orchestrates the response generation. It requests relevant context from the Retrieval Agent, constructs a detailed prompt, and queries the Google Gemini LLM to generate a final answer.

---

🚀 Getting Started

Follow these steps to set up and run the project on your local machine.

1. Prerequisites

• Python 3.10 or higher
• pip package manager
• git (optional, for cloning)

2. Clone the Repository

First, get the project files onto your local machine.

git clone https://github.com/your-username/agentic-rag-chatbot.git
cd agentic-rag-chatbot

3. Create a Python Virtual Environment

It is highly recommended to use a virtual environment to manage project dependencies and avoid conflicts with other Python projects.

On macOS/Linux:

python3 -m venv virtual_container

On Windows:

python -m venv virtual_container

4. Activate the Virtual Environment

Before installing dependencies, you must activate the environment.

On macOS/Linux:

source virtual_container/bin/activate

On Windows:

.\virtual_container\Scripts\activate

Your terminal prompt should now be prefixed with (virtual_container), indicating that the virtual environment is active.

5. Install Dependencies

Install all the required Python packages using the requirements.txt file.

pip install -r requirements.txt

6. Set Up Google API Key

This project uses the Google Gemini API for language model capabilities. You need to provide an API key as an environment variable.

1. Obtain a free API key from Google AI Studio.
2. Set the environment variable.

On macOS/Linux:

export GOOGLE_API_KEY="YOUR_API_KEY_HERE"

To make this permanent, add the line above to your shell's configuration file (e.g., ~/.bashrc, ~/.zshrc).

On Windows (Command Prompt):

set GOOGLE_API_KEY="YOUR_API_KEY_HERE"

On Windows (PowerShell):

$env:GOOGLE_API_KEY="YOUR_API_KEY_HERE"

Note: If you do not provide an API key, the application will run in a fallback mode, providing responses based on direct context snippets instead of a generated summary.

7. Run the Application

Once the setup is complete, you can start the Streamlit application.

python3.10 -m streamlit run main.py

The application will automatically open in your web browser, typically at http://localhost:8501.

---

📖 How to Use

1. Upload Documents: Use the sidebar to upload one or more supported files (pdf, pptx, csv, docx, txt, md).
2. Wait for Processing: The files will be processed and ingested into the vector store. You'll see a success message for each file.
3. Ask Questions: Type your questions into the chat input at the bottom of the page and press Enter.
4. View Responses: The chatbot will respond with an answer based on the information in your documents.
5. Check Sources: You can expand the "Sources" section below each answer to see which documents (and specific parts) were used to generate the response.

Project Structure

agentic-rag-chatbot/
├── requirements.txt
├── main.py
├── agents/
│   ├── __init__.py
│   ├── base_agent.py
│   ├── ingestion_agent.py
│   ├── retrieval_agent.py
│   └── llm_response_agent.py
├── core/
│   ├── __init__.py
│   ├── mcp.py
│   ├── document_parser.py
│   └── vector_store.py
├── ui/
│   ├── __init__.py
│   └── streamlit_app.py
└── config/
    ├── __init__.py
    └── settings.py

⚙️ Key Dependencies

• streamlit: For creating the interactive web UI.
• chromadb: The vector database for storing and searching document embeddings.
• sentence-transformers: For generating high-quality semantic embeddings of text.
• google-generativeai: The official Python client for the Google Gemini API.
• PyPDF2, python-pptx, pandas, python-docx: For parsing various document formats.