KNN vector search engine
A custom KNN search engine that ingested more than 12 million records at 10,000 documents a second and retrieved in under 100 ms.
I built a KNN vector search engine for candidate recommendations at GoodSpace AI. It ingested more than 12 million records at 10,000 documents per second and returned results in under 100 ms.
Retrieval quality came from custom weighting logic on top of the vector search, so the matches were relevant and not only near.
- Records
- 12M+
- Ingestion
- 10,000 docs a second
- Retrieval
- under 100 ms