User specific retrieval


I am curious at which level to handle filtering user-related vector embeddings.

In more detail, I want to enable document retrieval and q-and-a for each user instance. I wondered if I should initialize the documents into a vector database and then filter the vector by the metadata e.g. user_id when I want to retrieve similar documents.

My goal is to incorporate a lot of document embeddings per user and have the best possible performance therefore, would like to know about the best practices on how to store a lot of different documents (which should not be exposed to other users) in pinecone.

Thank you for your feedback.

Hi @lukasdavidebeling ,

Thank you for the interesting question!

Your use case would perfectly match a Pinecone feature known as namespaces.

Namespaces are like partitions, and they are ideal and completely appropriate for storing separate user data in the exact manner you’re describing.

Here’s a link to the documentation for namespaces. When you query a user’s embeddings, you’ll pass the unique namespace value that identifies them.

I hope that helps, and let me know if anything is unclear!