This article explores integrating AWS Responsible AI dimensions into customer support chatbots built with Amazon Aurora PostgreSQL. It details a design pattern for the Explainability dimension, demonstrating how to use RAG with vector embeddings and database-layer auditing to provide grounded, verifiable AI responses.
Key points
Log retrieval provenance, including chunk IDs, similarity scores, and guardrail traces, directly in the database for auditability.
Maintain a clear distinction between retrieved context and the specific sources cited in a response to ensure explainability.
Use Row-Level Security (RLS) in PostgreSQL to enforce tenant isolation before executing vector searches.
Treat audit and log tables with the same security and governance protocols as the primary source data.