This looks super neat, but only works on generated PDFs, not images, for now. While it uses imaging tricks to extract lines and table-like structures, it needs

This looks super neat, but only works on generated PDFs, not images, for now. While it uses imaging tricks to extract lines and table-like structures, it needs actual text (not images of text) to extract. So, there's an OCR step missing. Something that could take a picture of a bill or invoice and structure that data would be pretty useful, I imagine (even better if it ran on mobile devices). Even something that used a third party (like AWS Rekognition, whose text bits are pretty great) and added text would help this library be useful in more cases, without really having to add features to it.