RAG Engineer skill
Explore the open GitHub skill for tracing a RAG system from source knowledge to retrieval, grounded answers, and verifiable citations.
AI Engineering Lab · Open methodology
Build AI assistants that answer from trusted knowledge.
RAG Engineer is a portable workflow for building, debugging, reviewing, and operating retrieval-augmented generation systems. It traces evidence from the original source to the final cited answer.
What it covers
Fix the first broken stage.
A better prompt cannot recover a fact that was never indexed, retrieved, authorized, or preserved in the final context.
Debug each pipeline stage
Trace source, indexing, authorization, retrieval, assembly, generation, citation, and evaluation in order.
Enforce the access boundary
Resolve access on the server before retrieval and keep unauthorized documents out of raw results.
Operate with evidence
Keep quality, refusal behavior, latency, cost, outages, re-indexing, and deletion consistency measurable.
Related method
Prove that it works.
Pair RAG Engineer with RAG Evaluation to separate retrieval failures from generation failures and support defensible quality claims.