Main Concept

  • You connect the model to your external knowledge base (you own data)
  • External data could be documents, databases, wikis, etc.
  • The external data is pre-indexed and controlled (you decide what information the model can access)
  • You external data is converted to vectors (embeddings), and stored in databases like Amazon OpeSearch
  • The model retrieves relevant information, then generate responses.
  • Example: Customer service chat for a company
    • Knowledge base: Company’s products manuals, FAQ documents, policy guides stored in Amazon S3.
    • Process: Customers asks, the system searches on the company documents and generates answers.