Skip to content

RAG Reference Implementation

A production-pattern RAG application built end-to-end with Claude — from architecture decisions and cost modeling through to tested, deployed code.

~$8–13Monthly cost, per env at 1k queries
19ADRs
7C4 Architecture Docs
AWSEnd-to-end Deployment
CI IntegratedCost Modeling

What this is

A reference implementation of a Retrieval-Augmented Generation system, built to demonstrate production-pattern AI application delivery using Claude as the primary delivery agent. It answers questions grounded in a Hugging Face dataset, citing the source chunks it used. Every architectural decision is documented and traceable; every PR is reviewed against org-wide standards enforced in CI.

Cost-conscious

Scale-to-zero on AWS: Aurora Serverless v2 (0-ACU auto-pause), Lambda, and pay-per-token Bedrock. Estimated $8–13/mo per environment at 1k queries ($16–26 for QA + Prod).

Git-native docs

This site (Astro Starlight) embeds the LikeC4 architecture model, versioned with the code.

Cost in CI

Terraform with an Infracost monthly-cost diff posted on every pull request.

Safe

Tech currency baked into CI, and safety baked into the design.

Runs locally

Postgres + pgvector in Docker, FastAPI, and a React UI. Bedrock calls hit real AWS with your credentials.