Design responsible AI controls for Amazon Aurora PostgreSQL
Machine Learning Blog
This article demonstrates how to design responsible AI controls for a customer support chatbot built on Amazon Aurora PostgreSQL using Retrieval Augmented Generation (RAG) with pgvector embeddings and Aurora ML.
- Maps eight AWS Responsible AI dimensions (Fairness, Explainability, Privacy, Safety, Controllability, Veracity, Governance, Transparency) to Aurora PostgreSQL capabilities
- Implements Explainability in depth through comprehensive interaction logging: retrieved chunk IDs, similarity scores, cited sources, and guardrail traces
- Uses Aurora ML to generate vector embeddings directly from SQL without data movement or custom ETL pipelines
- Enforces tenant isolation with Row-Level Security and redacts PII before embedding using Amazon Bedrock Guardrails
- Provides database schema design with chat_interactions table capturing full RAG provenance for audit trails
- Includes pgaudit and pg_cron for governance, audit logging, and automated 90-day retention with S3 archival
- Demonstrates contextual grounding with Bedrock Guardrails to verify responses are supported by retrieved sources
The design patterns enable queryable audit trails at the database layer, distinguishing cited from retrieved sources to make AI responses fully explainable and traceable.
The AWS News Feed is currently looking for gold sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.
Related articles
2026
2026
2026
2026
The AWS News Feed is currently looking for silver sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.