bedrock-rag-qa-assistant

Bedrock RAG QA Assistant is a Retrieval-Augmented Generation (RAG) Q&A service built on Amazon Bedrock. It uses Titan Embeddings for retrieval, generation via Bedrock's Converse API, ChromaDB for vector storage, and offers a serverless deployment path via Lambda and API Gateway with Terraform. The project is designed so users can ask a question and get an answer grounded in their own documents, with sources cited, either running locally or deployed serverless on AWS.

The repository provides two deployment shapes using the same code. A local/API shape runs a FastAPI app with uvicorn, while a serverless shape wraps the same FastAPI app for Lambda via Mangum, fronts it with API Gateway, and provisions resources with Terraform, staging the deployment package through S3. The tech stack includes Amazon Bedrock Titan Embed Text v2 for embeddings, Bedrock's Converse API for provider-agnostic generation across Claude or Nova, ChromaDB as the vector store, FastAPI + Uvicorn for the API, AWS Lambda via Mangum + API Gateway for serverless runtime, and Terraform for infrastructure as code.

The project demonstrates Amazon Bedrock embeddings and generation, RAG pipeline design, prompt engineering, Infrastructure-as-Code for a serverless AWS deployment with IAM scoped to Bedrock inference-profile ARNs, real deployment troubleshooting, API design, automated testing, and CI. It includes a Streamlit demo UI, pytest tests with mocked Bedrock calls, and documentation covering architecture, quickstart, and API reference. A roadmap lists swapping ChromaDB for OpenSearch Serverless, adding streaming responses, adding re-ranking, multi-format ingestion, and API key/usage-plan on the API Gateway route.