For vector search I noticed most of the names I’ve heard about in the past missing. Faiss, Annoy, hnswlib, vespa. Hell, if you have less than a a million vectors you can probably even just use the sentence transformers lib: <a href="https://www.sbert.net/examples/applications/semantic-search/README.html" rel="nofollow noreferrer">https://www.sbert.net/examples/applications/semantic-search/...</a><p>Optimize for retrieval metrics first, then speed. Recall and coverage. If those suck, it’s because your embeddings are trash and it won’t matter how fast you serve them.