Skip to main content
Technical implementation · AI Search Infrastructure

​
Definition

Approximate nearest neighbor search is the algorithm that finds the vectors closest to a query vector in a large index. It trades a small, controlled amount of recall for dramatic speed gains over exact search. Exact search over millions of vectors is computationally prohibitive at production scale. ANN algorithms — HNSW, IVF, and others — make vector retrieval fast enough for real-time query responses. The approximation means that a small percentage of genuinely relevant chunks may not appear in first-pass results. This is why the reranking stage exists: to catch precision failures that the approximate first pass missed.

First-pass retrieval

Reranking

Vector database

Embedding

Dense retrieval

​
Relevant Plate Lunch Collective Services

AI SEO