Question

A financial services company is building a RAG application on Amazon Bedrock. They have created a Bedrock Knowledge Base backed by an OpenSearch Serverless vector store. Their compliance team requires that all retrieved document chunks be logged and passed through an internal policy-validation microservice before any text is sent to a foundation model for generation. The team also wants full control over the final prompt template and to potentially route different query types to different models. Which approach best meets these requirements while still leveraging the managed Knowledge Base?

Accepted Answer

Use the Retrieve API to fetch relevant chunks from the Knowledge Base, run them through the policy-validation microservice, then construct the prompt and call the InvokeModel API separately

Answer

Use the RetrieveAndGenerate API and configure a Lambda post-processing hook to inspect the generated answer before returning it to the user

Answer

Use the RetrieveAndGenerate API with a custom prompt template and rely on Bedrock Guardrails to enforce all compliance policies during generation

Answer

Disable the managed Knowledge Base and implement a custom retrieval pipeline that queries OpenSearch Serverless directly, then call InvokeModel