Whilst this is cool, why would we want to push this logic into the DB?

It seems cleaner to keep this in the service layer and use any PDF parsing library and subsequent schema to store the parsed files.