Exactly, I think that in this case using ML models that are experts in papers can be more useful and that type of models can run locally so it's the best option.