[simonw]

That does look pretty great, thanks for the tip.

Sending images through that API and then using an LLM to extract data from the text result from the OCR could be worth exploring.