Retrieval & RAGintermediate12 min · 9 steps
RAG Architecture Variants
Beyond one query and one retriever: expand the query, run many variants, fuse many retrievers, search a document hierarchy top-down, and refine candidates through a multi-stage funnel.
recall@1062%
retrieval60ms
requesterrordata
metrics are illustrativeProblemStep 1 of 9
One phrasing, one retriever, one chance
Standard RAG embeds the query once and searches one index. If the user said “heart attack” and the guidelines say “myocardial infarction”, or the answer sits in clause 7.3.2 of a 300-page manual, that single shot misses.
The variants in this lesson attack the recall ceiling — the relevant chunk never making it into the candidates — by multiplying queries, retrievers or levels.
Lesson 26 of 37 in Retrieval & RAG