Advanced features

Use this section after ordinary indexing and searching work. Read the result-shaping or vector subsection that matches your application requirement.

Result shaping

Feature	What it does	When to use it
Highlighting	Wraps matching terms in markup for display	Search result snippets
Field collapsing	Groups results by a field value (e.g. one hit per category)	Deduplicating search results
Aggregations	Computes min, max, sum, count, average, and histograms over a numeric field	Analytics, dashboards, faceted counts
Faceting	Counts matching documents by DocValues field	Navigation, filters, and category summaries
Reciprocal rank fusion	Merges multiple query result sets without score calibration	Hybrid search (BM25 + vector)
Geo search	Bounding box and distance queries over latitude and longitude points	Location-based filtering

Vector search

Feature	What it does	When to use it
Vector search	Approximate nearest neighbour over dense float vectors using HNSW graphs	Semantic search, embeddings, similarity
Filtered vector search	Vector ANN with a pre-filter or post-filter query	Scoped semantic search

Vectors can be quantised with BBQ (Better Binary Quantisation) for 32x compression with minimal recall loss. The quantised query path operates in int8 for speed.

Specialised queries

Feature	What it does	When to use it
Block-join	Queries parent documents based on child matches	Nested documents (blog posts with comments)
More like this	Finds documents similar to a given document	Related content, recommendations
Spelling suggestions	Did-you-mean corrections based on the index lexicon	Search UX

Scoring and ranking

Feature	Where to learn	What it does
BM25+ and BM25L	Boosting and scoring	Advanced BM25 variants with lower-bound and length normalisation
Block-Max WAND	Search internals	Score-bound skipping for supported top-N queries
Language-model similarities	API reference for DirichletSimilarity, etc.	Probabilistic relevance models
SIMD cosine	Vector search	Vectorised cosine similarity for dense vectors