RAG — the model answers from what you retrieved, right or wrong
Retrieval-augmented generation grounds an answer in fetched documents instead of the model's memory, which moves the failure mode from hallucination to retrieval quality — you now fail by fetching the wrong passage.
Not yet scoped.
Why
The PoC builds a small RAG pipeline (chunk, embed, retrieve, answer) and measures answer quality against retrieval quality, showing the bottleneck is upstream of the model.
How it works
Not yet built.
Where it lands in Jayverse
Number: if Number ever adds search or an assistant over research notes, build the eval around retrieval precision, not answer fluency. This PoC's point is that the bottleneck is upstream of the model, so measure the retrieve step directly.
Auditor: wherever a service uses RAG (a Number assistant, a persona chat in Rabbit), track retrieval quality as its own metric, separate from answer quality, so a wrong-passage failure isn't mistaken for a model failure.
Key expressions
Words and phrases from this page worth keeping, with the Korean meaning and the sentence they come from.
Expression
뜻 · 쓰이는 자리
ground (동사)
(주장·답을) 근거에 기반하게 하다 · 모델 답변을 외부 문서에 근거하게 만들 때. "grounds an answer in fetched documents"
bottleneck
병목 · 성능을 가장 크게 제약하는 지점을 가리킬 때. "showing the bottleneck is upstream of the model"
upstream of
~보다 앞단계에·상류에 · 문제의 원인이 모델보다 이전 단계(검색)에 있을 때. "the bottleneck is upstream of the model"
fail by
~함으로써 실패하다 · 실패가 구체적으로 어떤 행동 때문에 생기는지 말할 때. "you now fail by fetching the wrong passage"
failure mode
실패 양상·고장 유형 · 시스템이 잘못될 때 나타나는 전형적인 방식. "moves the failure mode from hallucination to retrieval quality"
RAG
검색증강생성(Retrieval-Augmented Generation) · 모델이 외부 문서를 검색해 그 내용에 근거해 답을 생성하는 방식. "Retrieval-augmented generation grounds an answer in fetched documents"