PostrgreSQL with Pgvector extension or Astra DB as a vector DB for RAG approach using Bedrock?

Question

Hi!

I'm building a RAG approach app using bedrock behind the scenes. I was searching and testing different vectorial databases, but since I need to have more control over the data and filter it per user in my case I'm using PostgreSQL with pgvector. But doing some research I found out about Astra DB.

Which one is better for RAG approach and embeddings? Both are good options? Or One of them is better than the other one?

Thank you!

Accepted Answer

Hi,

I like this post which depicts well the advantages of all possible database types that you can use for embeddings: https://www.linkedin.com/pulse/choosing-vector-database-your-gen-ai-stack-abhinav-srivastava/

Personally, I would initially try with pgvector as it is available on AWS Aurora and see how far I go with it: https://aws.amazon.com/about-aws/whats-new/2023/07/amazon-aurora-postgresql-pgvector-vector-storage-similarity-search/

This post gives the limits of pgvector if you are at really high scale: https://medium.com/@zilliz_learn/getting-started-with-pgvector-a-guide-for-developers-exploring-vector-databases-9c2295bb13e5

Best,

Didier

PostrgreSQL with Pgvector extension or Astra DB as a vector DB for RAG approach using Bedrock?

Relevanter Inhalt