Ask HN: State of the art in document data extraction
I used to run a side-project which turned PDFs into data. It started as a bit of client work back in 2013 (Scala + PDFBox, with thousands of PDFs with the same layout needing to be parsed) and evolved over the years to be a general-purpose data extractor, with a focus on tabular data (with the help of a PhD computer vision expert, lots of deep learning + ML; up to 2018).
It was during 2018 when my PDF project got me head-hunted for a consultation gig to build a receipt scanner for a large UK customer data science company. The CTO suggested we use Google Vision, and it was quite easy to use the provided bounding boxes for paragraphs/sentences/words/letters to isolate rows/columns. When I revisited my own project after this gig, I threw a lot of my old code away, because Google Vision clever isolation of sections did all the heaviest lifting.
I'm curious to know if anyone else works in this field, what tools you use (e.g. Adobe PDF Extract, Google Document AI, etc), and what types of problems you're trying to solve.