Struct AnnIndex 

pub struct AnnIndex { /* private fields */ }

Approximate nearest neighbor index using HNSW for large datasets, with brute-force fallback for small ones.

Vectors are stored in a FlatEmbeddings — one contiguous Arc<[f32]> allocation — shared with the rest of the pipeline.

Implementations

impl AnnIndex

pub fn build(embeddings: FlatEmbeddings) -> Self

Build the index from a FlatEmbeddings.

The corpus is shared via FlatEmbeddings.data (an Arc::clone, zero bytes copied), so the cached HNSW index shares the same flat f32 allocation as the per-query aligned corpus.

pub fn search(&self, query: &[f32], top_k: usize) -> Vec<(usize, f32)>

Search for the top-k nearest neighbors of a query vector. Returns (index, similarity) pairs sorted by descending similarity.