Insurance Chatbot with Conversational RAG and VectorDB

Problem

Insurance agents have to provide instant, accurate responses to customer queries, but accessing relevant information quickly can be challenging.

Solution

The conversational RAG Bot with voice-enabled interaction addresses this by providing seamless, real-time assistance integrated into the frontend.

Tools

• LLM - Cohere’s Command-r-plus: Best suited for complex RAG workflows with long context.
• VectorDB - Qdrant Vector Store & ChromaDB: Useful for semantic-based matching.
• Document Loader: Unstructured text extraction.
• Uvicorn: ASGI web server.
• FastAPI: High-performance web framework for APIs.
• Axios: HTTP client for API requests.
• Browser API:
  • SpeechRecognition - Voice to Text conversion
  • SpeechSynthesis - Text to Speech conversion

Implementations

• Loaded 20+ Policy documents
• Conversational RAG Technique
  • Indexing
  • Retrieval (Chat History + Present user Query)
  • Generation
• Speech Recognition
  • Recognize voice input from users and convert it into text in real-time.
  • Enabling the chatbot to respond with audio output.

Project Outcome

Created a responsive, real-time chatbot with retrieval-based knowledge augmentation to answer user queries related to policy documents.

Chatbot Interface