A generic high-level "descriptor" (likely LLM-based) metadata extractor

[yarikoptic]

I wondered if anyone worked or came up already on some LLM based metadata extractor which would for an arbitrary dataset extract some description

Ultimate "goal" -- to add/enable on https://registry.datalad.org/ to be able to find data of interest (e.g. currently was looking for "geo-spatial" data)

Filed similar issue (potentially with more detail) elsewhere:

• Automated (LLM, agents, ...) metadata extractor/aggregator to describe directory (dataset) in datacite schema datacite/datacite-suggestions#185