v0.5.1 LLM-agnostic RAG pipeline

attune-rag

Lightweight, LLM-agnostic retrieval with measured quality — and CI that fails when the numbers drop.

pip install attune-rag

100%Precision@1

100%Recall@3

0.97Faithfulness

Bundled-corpus actuals, gated in CI — a PR that regresses them fails automatically. On an unseen corpus, the optional transformer tier lifts hard-paraphrase precision@1 from 25% → 90% and recall@3 to 100%.