[ericd]

I briefly tried using the results of that, they're totally unnormalized, so even lining up identical line item charges is difficult. We have a deep learning language model that normalizes the language and lines up potential matches. But even if that works, it still doesn't get you the bill, because a single encounter includes a big batch of billing codes, depending on what the doctor tries to give you. And to get normal baskets, you probably want claims data. And the claims data providers we spoke to wanted 6-7 figures for a small dataset.

It's an incredibly parasitic industry, I've never seen anything quite like it. It's a pretty stark contrast to the largely altruistic motives of many of the healthcare professionals themselves.