Live Multi-User RAG Chatbot

A real-time Retrieval-Augmented Generation (RAG) chatbot built to support multiple concurrent users using asynchronous message processing and semantic search.

Features

• Multi-user question answering with isolated chat memory

• Retrieval-augmented responses (no hallucinations)

• Asynchronous processing via RabbitMQ

• Fast semantic search using Pinecone

• Scalable, fault-tolerant architecture

Architecture

RabbitMQ decouples user requests from heavy RAG processing

Pinecone stores and retrieves vector embeddings

LLM generates grounded answers using retrieved context

RAG Workflow (Step-by-Step)

1. User asks a question

2. Request is sent to the API

3. API publishes a task to RabbitMQ

4. RAG worker consumes the message

5. Query is embedded and sent to Pinecone

6. Top relevant documents are retrieved

7. Prompt is constructed using retrieved context

8. LLM generates a grounded answer

9. Response is returned to the user