Thoughts

Can the documents grouped in any way before hand, or they could be about anything?

@Logan M They certainly can be. Here is a similar example that may be relatable but not exactly my use case.

Pretend we are indexing legislation & precedents. The legislation would be querying only 1 document at a time for an answer specifically relating to that law/act. Eg. How large can my shed be before I need council consent?

The second set would be precedents which if queried, would need to check multiple documents (can be further refined into sectors: eg. property, civil, etc. ) and then return an answer that could relate to multiple sources.

I think your onto the right track with the graph approach! You can set up some pretty complex graphs that would allow this

Have you seen the latest guide on graphs? It covers some pretty complex stuff as well, you might find it useful! https://gpt-index.readthedocs.io/en/latest/guides/tutorials/graph.html

@Logan M I appreciate your input! I did have a brief look over that yesterday but aiming to give it a bit more time tomorrow. Overall just wanted to make sure I haven't made a glaring error in logic from the start and then chasing down the incorrect rabbit hole.