hi, I want to build a gent to answer high-level research questions from 100 papers. In order to do it, I am thinking of building 100 tools, with each tool query each paper. Is there any suggestion how to write a good description of each tool? I am think of writing a prompt to automatically generate the metadata.
I also just found this blog gives a good summary of the challenge here: multi-hop question-answering. In my demo, I have observed that the top-k are usually coming from the same paper, and thus does not provide comprehensive answers.