Hacker News Reader

We use ScaNN for a large scale/performant neural search. Otherwise this all feels bloated.

Another upvote for ScaNN. Used on Teclis.com search engine in production (replaced previously used Faiss, which was excellent). ScaNN is almost double the performance and with same 5 lines of code to implement locally.

At this point vector search is almost 'solved', the challenge remains the quality of vectors fed into the index. Producing accurate text/image/... vector representations (aka embeddings) is where lot more progress needs to be made.

There are many "extra" features that one needs in practice which are patently not solved, starting with working in a dynamic setting.

> At this point vector search is almost 'solved'

Well, there is is still a lot to do. Going from memory to disk, etc. But I see your point.

What might be a bit confusing is the mix of vector search libraries with vector search engines. It's a bit like comparing an inverted index library with SOLR :)

But we (I.e., the wider vector search ecosystem) are working on this

What sort of scale are you dealing with? ScaNN looks more like a vector index like FAISS, rather than a full database.

Reply on news.ycombinator.com