Instead of having to have 27 columns you have one... I guess?

This also makes interop more difficult when working with indexing databases (i.e. elastic).

This does minimize the amount of client code required to parse pdf's though.

But then it means I need to save the PDF in the database rather than some kind of storage

psql is very good at that since 9.0

RAG! AFAICT it doesn’t do OCR off the shelf, but surely you could hook up Nougat/Tesseract/TOCR without too much difficulty and start parsing any PDFs that’s aren’t parsed yet.