Multi-tenant RAG Engine

This project aims to provide a unified vector database solution for multiple AI applications. By leveraging collections or namespaces within a single vector database instance, we can effectively separate data for different websites or applications.

Core Architecture

• Vector Database: A single container instance serving as the storage layer, using namespaces to maintain data isolation.
• Central Processing Engine: A central container responsible for:
  • Receiving user queries.
  • Orchestrating AI API calls.
  • Handling embedding model generation (converting website data to vector embeddings).
  • Communicating with the LLM to generate context-aware, accurate responses based on the retrieved data.

RAG Principle

This engine follows the Retrieval-Augmented Generation (RAG) principle. It retrieves relevant knowledge from the vector database for a specific tenant/website and feeds it to the LLM as context, ensuring answers are grounded in real-time, tenant-specific information.

Use Case: Starfolio Assistant Integration

The first challenge is integrating an AI assistant into the existing Starfolio website:

• UI Integration: Adding a floating chat button in the bottom-right corner that opens an overlay interface.
• Data Challenge: Transforming the pre-existing content of the Starfolio website into a format suitable for the vector database (embedding generation) so the assistant can answer questions based on your portfolio and blog content.