Proper way to setup RAG with Gumloop

Is there a guide or best practices on how to configure Gumloop to use with local documentation for AI RAG? It seems like I could upload the documents to Google Drive and then have Gumloop retrieve those during the flow, but I have more questions…

  1. If the documents are just stored in Google Drive for example, am I correct in thinking that there would be no vector database involved? Wouldn’t that slow things down and/or make the output less accurate?

  2. Can I route the users question (based on keywords) to a specific AI model and set of documents based on the input. For example, if the user asks about “cats”, but my Google Drive has documents on “cats” and “dogs”, can we automatically route the AI to search only the “cats” documents for the answer (and using a specific system prompt related just to cats)? If so, is there a proper way to setup the Google Drive (i.e. a folder for each subject)?

I am sure I will have more questions as I start to build this out, but I want to see if RAG is even recommended with Gumloop.

Hey @DougA - This could be possible with Gumloop however it’ll be very tricky to set up. I’d highly recommend creating a custom node that takes care of the RAG process rather than trying to string together a flow that does RAG.

This topic was solved and automatically closed 60 minutes after the last reply. New replies are no longer allowed.