Hardest AWS Generative AI Developer - Professional questions this week

A Knowledge Base backed by OpenSearch Serverless performs semantic retrieval over the S3 docs, and Haiku gives the lowest latency and cost among Bedrock models, fitting the budget. Guardrails' denied topics policy is the managed way to reliably suppress competitor mentions. Fine-tuning with provisioned throughput is costly overkill for grounding in existing docs. Pasting all documentation into the prompt blows past context limits and inflates per-request cost. Relying only on a system-prompt instruction to suppress competitor mentions is unreliable compared to a dedicated guardrail, and a relational lookup skips semantic search entirely.

A single Bedrock Agent with separate action groups provides clean separation of concerns. Each action group has its own Lambda function and OpenAPI schema, allowing the developer to implement identity verification in the transaction-related Lambda functions while keeping the report generation Lambda simple. This is the least operational overhead approach that properly enforces security at the execution layer.

Bedrock Guardrails' sensitive information filters support built-in PII types (including SSN) and custom regex patterns for organization-specific identifiers like medical record numbers. The denied topics policy handles blocking medical diagnoses and self-harm discussions. Content filters handle harmful content categories. Enabling CloudWatch logging on the guardrail captures all intervention events for compliance auditing. This configuration addresses all four requirements natively.