What changed
AWS has introduced a design pattern for implementing responsible AI controls in customer support chatbots built on Amazon Aurora PostgreSQL-Compatible Edition. The approach maps the eight AWS Responsible AI (RAI) dimensions to practical implementation strategies, with a focus on the Explainability dimension. This includes using Retrieval Augmented Generation (RAG) with pgvector embeddings, Amazon Bedrock model inference, and AWS Responsible AI guardrails to deliver grounded, explainable responses.
What this means in practice
The solution involves an ingestion pipeline that redacts personally identifiable information (PII) from support tickets, FAQs, and product documents before converting them into vector embeddings. When a user submits a query, AWS Lambda runs a semantic search over Aurora PostgreSQL, applies Amazon Bedrock Guardrails for privacy and grounding, sends the guarded context to an Amazon Bedrock foundation model (FM), and logs every interaction for explainability and governance. Aurora ML enables invoking Amazon SageMaker endpoints and Amazon Bedrock foundation models directly from SQL, without data movement or complex integrations. The chatbot can answer questions like 'Why was my last bill high?' or 'Where is my last order?' while maintaining privacy through PII redaction and access controls.
Limitations and availability
The solution is designed as part of a series and focuses on the Explainability dimension. It requires Aurora PostgreSQL-Compatible Edition with Aurora ML and the pgvector and aws_ml extensions for vector operations. The ingestion pipeline must redact PII before embedding, and the same Amazon Titan Text Embeddings V2 model must be used for both the knowledge base and user queries. The architecture includes data ingestion, query and response flow, and responsible AI controls applied at each stage. The solution is available for custom logging design when provenance needs to be stored in the user's own schema.
