RAG Comparison App

A Streamlit application that compares Traditional RAG (Retrieval-Augmented Generation) with Agentic RAG using CrewAI. This application allows users to upload PDF documents and ask questions about them while comparing the responses from both approaches.

Features

• PDF document processing and chunking
• Dual RAG implementation:
  • Traditional RAG using LangChain
  • Agentic RAG using CrewAI
• Semantic conversation history tracking
• Unified vector store for both PDF chunks and conversation history
• Response quality control using Guardrails
• Interaction tracking using Opik
• Side-by-side comparison of responses
• Citation tracking and display

Installation

1. Clone the repository:

git clone https://github.com/Praneeth16/rag-comparison-app.git
cd rag-comparison-app

2. Create and activate a virtual environment:

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

4. Install dependencies:

pip install -r requirements.txt

5. Create a .env file in the project root and add your API keys:

OPENAI_API_KEY=your_openai_api_key
OPIK_API_KEY=your_opik_api_key

Usage

1. Start the Streamlit app:

streamlit run app.py

2. Upload a PDF document using the sidebar

3. Ask questions about the document using the chat interface

4. Compare responses from both RAG implementations side by side

Project Structure

rag-comparison-app/
├── app.py                      # Main Streamlit application
├── requirements.txt            # Project dependencies
├── .env                       # Environment variables
├── components/
│   ├── __init__.py
│   ├── pdf_processor.py       # PDF processing utilities
│   ├── vector_store.py        # Unified vector store manager
│   ├── rag_handler.py         # Traditional RAG implementation
│   ├── agentic_rag_handler.py # Agentic RAG implementation
│   ├── tracking.py            # Interaction tracking
│   └── guardrails.py          # Response quality control
└── vector_store_db/           # Vector store database directory

Features in Detail

Traditional RAG

• Uses LangChain's ConversationalRetrievalChain
• Implements semantic search for relevant document chunks
• Maintains conversation history for context
• Provides source citations

Agentic RAG

• Uses CrewAI for agent-based interactions
• Implements three specialized agents:
  • Context Analyzer: Understands conversation history
  • Research Analyst: Finds relevant information
  • Technical Writer: Creates comprehensive responses
• Provides detailed citations and explanations

Vector Store

• Unified ChromaDB instance for both PDF chunks and conversations
• Metadata-based filtering for different types of content
• Semantic search capabilities
• Persistent storage

Quality Control

• Guardrails implementation to prevent hallucinations
• Strict context adherence
• Citation requirements
• Response validation