From the source
Amazon's guide maps the eight AWS Responsible AI dimensions to Aurora PostgreSQL features and builds out one of them, Explainability, for a RAG customer support chatbot.
- The reference design keeps the embedding pipeline inside the database, using Aurora ML to call Amazon Bedrock models from SQL without data movement or custom ETL pipelines.
- An ingestion pipeline redacts personally identifiable information from support tickets, FAQs and product documents, then converts the text into vectors with the aws_bedrock.invoke_model_get_embeddings function and Amazon Titan Text Embeddings V2 at 1,024 dimensions.
- At query time, Lambda embeds the question with the same model, runs a pgvector cosine distance search, applies Amazon Bedrock Guardrails for privacy and grounding, and sends the guarded context to a foundation model.
- Row-Level Security limits retrieved chunks to the current tenant, and every interaction is logged with citations and audit records for explainability and governance.
- The guide says Amazon Bedrock Knowledge Bases is the simpler choice when default retrieval and citation behavior meets your needs, and recommends the custom logging design when provenance must live in your own schema.
- It presents the Explainability work as Part 1 of a series rather than a full build-out of all eight dimensions.






