We had also considered creating a module that takes a bioformats file set as input, extracts useful structural metadata, and outputs a json/yaml/etc. file containing that metadata for the pipeline to use in various ways.

sounds interesting, perhaps a combination of Bio-formats and a LLMs could extract metadata and format it in a json or yml

I don’t think we need to bring LLMs into it

lol, true, thats probably overkill, but they can be quiet useful as agents