[x] Set up pipeline to run on ".pdf" ending documents, produce matches to reference & section.
[x] Download non-ending .pdf, then same.
[x] Complement it with contexts (ie. match first sentence), as there are instances where parser does not find all (ie. indirect citation in "the advent of advanced artificial neural network architectures [1–3] " for AlphaFold which is cited as [3] but only has that one reference. (and this one "y the Kabsch algorithm [3, 4]")
[x] Take the labelled data from ParsCit to classify headings from scipdf to them, with a category being miscellaneous.