Search code examples
chatgpt-apivector-databaseqdrant

Updating embeddings in vector databases

I'm new to how vector databases operate and the underlying concepts, especially when it comes to updating. I'm trying to apply the RAG pattern by combining a self hosted Qdrant database along with ChatGPT API. I'm creating embeddings by executing API calls to ChatGPT and saving them in Qdrant. Let's say I supplied contant of a document that contained proprietary information that was incorrect. How do I update my embeddings. If I only supply it with updated content, it would then contain contradictory information. Do I have to delete the whole collection and recreate it ?

Solution

  • You are correct. You will need to do two things:

    1. You need to delete the outdated content/collection.
    2. Insert the new content with its [new] associated embedding.