Launching today

brinicle
Extremely fast/RAM-friendly search engine
4 followers
Extremely fast/RAM-friendly search engine
4 followers
brinicle is a disk-first HNSW retrieval engine for vector search, structured item search, hybrid search, and autocomplete. On 1.2M Amazon products, brinicle achieved sub-ms P99 hybrid search (lexical, semantic) while using substantially less search memory than Weaviate, OpenSearch, Typesense, and Meilisearch. The report: https://brinicle.bicardinal.com/search_benchmark



I wanted to design an extremely fast vector search engine that does not blow your RAM, while performs competitive in terms of accuracy. So, I built brinicle.
brinicle supports:
ANN Vector Engine.
Item Search Engine.
And Autocomplete Engine.
I compared it against Milvus, Chroma, Qdrant, and Weaviate in different datasets and published results here: brinicle.bicardinal.com/benchmark
Then, I created a benchmark for its item search ability on two datasets and compared it against Weaviate, OpenSearch, Typesense, and Meilisearch. brinicle outperforms in terms of search latency, and memory consumption, while keeping the accuracy, or outperforming in some metrics. Results, and the approach are explained in this comprehensive report: brinicle.bicardinal.com/search_benchmark