Hey

I was following your Bottoms up tutorials on YouTube and I had a couple of questions. So I am doing a POC using a set of docs files that are in markdown. I have close to 3300 files in a nested folder structure. My question is:

1. At the top level I have 10 folders which broadly classify the category in which the document lies and inside of these 10 folders is a nested folder structure that contains these markdows. In terms of providing better context, does it make sense to delete the nested folder structure inside the 10 folders and have all the markdowns flatly available.
2. You tested the accuracy of your query engine using GPT4, is there any other way that I can test it out since I don't have access to it right now?
3. Are there any best practices that you recommend to have a better response accuracy.

I built my query engine locally which took around 2 hrs. Now when I send in a query, it takes roughly the same amount of time to generate the response. Is this behaviour correct or am I missing something ?

Is there a compatibility matrix indicating what version of langchain is supported with a particular version of llama index. I'm running into tons of dependency issues and it's driving me nuts just to go back and forth with the imports.