Project Overview
Situation
Enterprise RAG needs more than embeddings and a prompt — it requires authenticated access, department-scoped retrieval, reranking, guardrails, citations, and observability to be production-viable.
Task
Design a complete enterprise retrieval-and-generation architecture covering security, retrieval quality, response safety, and usage tracking.
Action
Architected the full pipeline: Microsoft Entra ID authentication → API request → configuration lookup → metadata/department filter building → AWS Bedrock retrieval → OpenSearch hybrid/vector search → reranking → generation → guardrail checks → citation extraction → usage metrics; designed for FastAPI + LangGraph orchestration with an Angular 21 frontend on AWS serverless (Lambda, API Gateway, S3, DynamoDB).
Result
A governed, citation-backed enterprise GenAI architecture applying production patterns — hybrid retrieval, multi-model routing, guardrails, and telemetry — end to end.