Comment by storus
8 hours ago
Pinecone allows hybrid search, merging dense and sparse vector embeddings that Postgres can't do AFAIK. That results in ~10% worse retrieval scores which might be the difference between making it in the business or not.
8 hours ago
Pinecone allows hybrid search, merging dense and sparse vector embeddings that Postgres can't do AFAIK. That results in ~10% worse retrieval scores which might be the difference between making it in the business or not.
The lesson for me isn't "don't use Pinecone", but more like "did you already max out Postgres?"
In many cases, it is going to save you time by having less infra and one less risk while you're getting started. And if you find yourself outgrowing the capabilities of Pg then you look for an alternative.
The article shows an example of hybrid search using RRF.
With BM25 which has a far worse/non-generalizable performance than sparse embeddings Pinecone supports. Moreover you get a latency hit from RRF that makes it challenging to use for e.g. real-time multimodal chat agents.