Car Insurance RAG Application

A Retrieval-Augmented Generation (RAG) application designed to answer questions about car insurance based on a provided FAQ document.

Overview

This project implements a QA system that uses semantic search to retrieve relevant information from a car insurance FAQ and then generates a detailed answer using a Large Language Model (LLM).

Key Features

• Semantic Search: Utilizes sentence-transformers/all-MiniLM-L6-v2 to create vector embeddings of the FAQ content.
• Efficient Retrieval: Uses FAISS (Facebook AI Similarity Search) for fast and accurate similarity searches.
• LLM Integration: Integrates with the Vextapp API (using Gemini 2.0 Flash) for generating human-like responses based on the retrieved context.
• Local Caching: Embeddings and the FAISS index are cached locally (paragraph_embeddings.npy and faiss_index.bin) to speed up subsequent runs.
• Interactive CLI: Simple command-line interface for user queries.

Prerequisites

• Python 3.7+
• A Vextapp API Key and Channel Token.

Installation

1. Clone the repository.

2. Install the required dependencies:

   pip install -r requirements.txt

Configuration

The application requires two environment variables to interact with the Vextapp API:

• VEXTAPP_CHANNEL_TOKEN: Your Vextapp channel token.
• VEXTAPP_API_KEY: Your Vextapp API key.

You can set these in your terminal:

export VEXTAPP_CHANNEL_TOKEN='your_channel_token'
export VEXTAPP_API_KEY='your_api_key'

Usage

Run the application using:

python rag_car_insurance.py

Once the system is ready, you can ask questions like:

• "What is third-party insurance?"
• "How are premiums calculated?"
• "What is a No Claim Bonus (NCB)?"

Type exit to quit the application.

Project Structure

• rag_car_insurance.py: Main application script.
• car_insurance_faq.txt: Source document containing FAQ paragraphs.
• requirements.txt: Python dependencies.
• faiss_index.bin: (Generated) FAISS index for vector search.
• paragraph_embeddings.npy: (Generated) Cached embeddings of the FAQ paragraphs.

Contact

If you have any queries, feedback, or suggestions, feel free to drop a mail at chetan.mahale0220@gmail.com :)