What's next in pgvector: Building AI-enabled apps with PostgreSQL - AWS Databases in 15

Generative AI and Large Language Models (LLMs) are powerful technologies for building applications with richer and more personalized user experiences. This year, Amazon Aurora and Amazon RDS announced support for pgvector, an extension that allows you to store embeddings from machine learning (ML) models in your database and to perform efficient similarity searches. In this session, we’ll briefly introduce you to pgvector, how it works, and why it’s important. AWS has had the opportunity to collaborate with the PostgreSQL community on this extension, we want to dive into some upcoming enhancements to pgvector. We will close by talking about how you can get started contributing and/or using this extension in your own workloads.



#AWS

[erikwiberg]

Thank for the video! As I'm understanding it, attaching metadata to the vector can result in a real performance boost when querying. Does pgvector support vector queries that also check against the metadata to narrow the search, or will that have to be broken up into separate steps manually in code? I can't seem to find clear documentation about this, about metadata size limit, supported query operators etc. I think Pinecone has a mongodb-like query language to match meta data like { metaDataKeyWithArrayData : { $includes: "my needle" } }. Is there something similar for pgvector?

[jdoejdoe]

Thank you for this presentation. Is there a way we can plug in an existing large postgress DB with multiple tables and perform complex queries using a natural language?