Definition
A retrieval architecture that sequences multiple search strategies, typically utilizing a fast, high-recall initial search (like BM25 or Vector Search) followed by a high-precision re-ranking stage using computationally expensive models to optimize context relevance.
Distinct from simple single-pass retrieval; it involves a 'coarse-to-fine' filtering process to balance latency and accuracy.
"A gold mining operation: a wide sluice box catches a large volume of sediment (initial retrieval), which is then hand-inspected under a magnifying glass to identify actual gold (re-ranking)."
Conceptual Overview
A retrieval architecture that sequences multiple search strategies, typically utilizing a fast, high-recall initial search (like BM25 or Vector Search) followed by a high-precision re-ranking stage using computationally expensive models to optimize context relevance.
Disambiguation
Distinct from simple single-pass retrieval; it involves a 'coarse-to-fine' filtering process to balance latency and accuracy.
Visual Analog
A gold mining operation: a wide sluice box catches a large volume of sediment (initial retrieval), which is then hand-inspected under a magnifying glass to identify actual gold (re-ranking).