#community-help

Understanding Vector Search with Typesense

TLDR In a chat with em1nos and Andrew, Kishore Nallan explained how Vector Search works. He clarified that it can be useful for recommendations and personalization, but it requires machine learning to convert data into vectors before searching.

Powered by Struct AI
Sep 11, 2022 (13 months ago)
em1nos
Photo of md5-f33ae8b829901656b655c985c894be3a
em1nos
08:01 AM
I just listened to Jason on The Changelog podcast.
Could you tell us a bit more about Vector Search? Sounds amazing!
How does it work?
What can we expect to be able to do, that we can't do right now with Typesense?
Is the RC publicly available?
I'm guessing the release of Vector Search is a few versions away? We should not expect it soon?
Kishore Nallan
Photo of md5-4e872368b2b2668460205b409e95c2ea
Kishore Nallan
10:11 AM
👋 em1nos

We've a RC build available already where you can try it out: typesense/typesense:0.24.0.rcn12

https://gist.github.com/kishorenc/f008c3a60ee58cb084b0c33c0dbce148

> What can we expect to be able to do, that we can't do right now with Typesense?
It allows you to do things like recommendations by building an embedding (which are vectors) which can be indexed into Typesense and searched using a vector query for fetching recommended items.

Still need to work on the docs that show how you can build a model that produces vector embeddings etc. if you aren't familiar with that.
em1nos
Photo of md5-f33ae8b829901656b655c985c894be3a
em1nos
10:15 AM
That's so interesting, but way over my head for now! I hope you guys can make this easy to work with.
Kishore Nallan
Photo of md5-4e872368b2b2668460205b409e95c2ea
Kishore Nallan
10:16 AM
That's the goal. We are just starting with the foundational stuff. Do you have any use cases in mind?
em1nos
Photo of md5-f33ae8b829901656b655c985c894be3a
em1nos
10:19 AM
Well, I'm actually using Typesense as part of a "find product matches" pipeline where I'm trying to find matching products within two datasets. I'm not sure if vector search can be of any use in that application.
10:19
em1nos
10:19 AM
I'm running a product/price comparison website, so I'm sure vector search will be of use in some way somewhere in the application.
Kishore Nallan
Photo of md5-4e872368b2b2668460205b409e95c2ea
Kishore Nallan
10:20 AM
Vector search works for recommendations and personalization.
em1nos
Photo of md5-f33ae8b829901656b655c985c894be3a
em1nos
10:22 AM
OK. I see. Maybe I got it wrong then what vector search actually is 🙂 I was thinking AI/ML stuff "more in general", and that it could be useful even outside of recommendations and personalization.
02:04
em1nos
02:04 PM
I guess vector search IS useful in many different contexts. We just need to understand it to really be able to make use of it.
For anyone interested in reading more about vector search:
https://www.infoworld.com/article/3634357/what-is-vector-search-better-search-through-ai.html
Or just google it, I guess.
02:04
em1nos
02:04 PM
My use case would be "deduplication" which is mentioned in that article.
Sep 12, 2022 (13 months ago)
Andrew
Photo of md5-88d88db4789daa0e3abef8c3ca27772b
Andrew
05:03 PM
Hey Kishore Nallan this sounds mega awesome. Will it work in foreign languages, in our case German?
05:03
Andrew
05:03 PM
We have been thinking how to find similar court judgements
Sep 13, 2022 (13 months ago)
Kishore Nallan
Photo of md5-4e872368b2b2668460205b409e95c2ea
Kishore Nallan
05:46 AM
What we are beginning with is the ability to index vectors and then do nearest neighbor search on them. So you still need to use a machine learning toolkit to convert the judgements into vectors instead of text but once you do that you can find similar court judgments by doing a nearest neighbor vector search.
Andrew
Photo of md5-88d88db4789daa0e3abef8c3ca27772b
Andrew
06:23 PM
Got it. Thanks Kishore Nallan