Best approach for multi-label multi-class texts in 2022?

I have more than 200k labeled text, all of them labeled in 2 Main-classes an every class has 200 sub-class , some of texts belongs to more than 3 classes ( 1 main-class and 2 sub-classes )

I’m going to give my model new text, my model should:
first) filter classes by some few questions at first,
second) guess what is topic of new text
(( its sort of topic modeling, multi-label multi-class classification, question answering method ))

what is best approach to do that based on new nlp Models?