Amazon's FinTech teams built a regulatory inquiry automation system on AWS using Amazon Bedrock, RAG with Knowledge Bases, and OpenSearch Serverless for vector storage. The architecture features a document ingestion pipeline supporting PDF, PPT, Word, and CSV formats with hierarchical chunking, a real-time WebSocket chat interface powered by Claude Sonnet 4.5 via the ConverseStream API, multi-turn conversation state managed in DynamoDB, and query expansion using Claude 3.5 Haiku to handle acronyms and vague queries. Parallel retrieval calls reduced latency from 10 seconds to under 2 seconds. Amazon Bedrock Guardrails filter PII and financial data, while OpenTelemetry integrated with a self-hosted Langfuse instance provides end-to-end observability for continuous improvement and compliance.