Uses Sato (repo, preprint) to identify the semantic type of columns.
I had to clean up their code to make it usable as a library here: github (mirrored from gitlab for convenience). I want to put this on PyPI, but because of the size of the model this might not be a great idea.