Advancing AI - Databricks Vector Search Index

The RAG (Retrieval Augmented Generation) pattern for keeping LLM's honest and accurate is super popular and being widely adopted, but you generally need to set up embeddings inside a Vector Database to get it working. Databricks recently released the Vector Search Index to automate this process for you, taking an existing Delta table and managing an underlying Vector Store!
In this video, Simon & Gavi look at the new Vector Search Index (VSI) functionality within Databricks, the limitations with the preview and the steps to get started working with it. Building a GenAI App in Databricks? This is your first step.
For more info on Databricks VSI, check out the docs here: docs.databricks.com/en/genera...
As always - if you're embarking on a GenAI application, get in touch with AA to give you a boost ahead!

[@christianw3858]

I see the benefit of the automated index update once you are pushing something to the table and not having to chunk that first. On the other hand, I see there a big disadvantage if you are applying a RAG architecture and feeding e.g. Azure OpenAI with it, as it returns the whole document rather than just the relevant chunks based on the vector search which can exceed tokens quite easily. I still would chunk it beforehand, do the embedding and udate the table. Do you know if this is possible and I would like to get the opinion on my statement here as well!

[@user-cz7yr9hs6g]

@user-cz7yr9hs6g 0 seconds ago Hello Simon. I have been a fan of your videos for a while. I wonder if you could answer this quick question. If we were to pursue using Databricks for an end to end RAG implementation , what is a good pattern that you may have seen, integrating Databricks with some sort of UI? Streamlit, is one the suggestions that I saw. But ideally, would love to have something that could be not a custom build.