- Datatypes to efficiently store floating-point and boolean vectors in Elasticsearch documents.
- Exact nearest neighbor queries for five similarity functions: L1, L2, Angular, Jaccard, and Hamming.
- Approximate nearest neighbor queries using Locality Sensitive Hashing and related algorithms for all five similarity functions.
- Combine nearest neighbor queries with standard Elasticsearch queries.
- Implemented using standard Elasticsearch and Lucene constructs, so indexing and queries scale horizontally with Elasticsearch.
- Horizontally scalable nearest neighbor search
- Visual similarity search
- Word and document embedding search
Elastiknn is very much a work in progress. I appreciate any feedback over on the Github repo.