I even went as far as building a small mobile app using AWS OCR. I was able to itemize and categorize using that but it still required manual "fingerprinting" of each receipt type. Basically figuring out that something was a Costco receipt and in Costco receipts this column in the OCR'd text was the name of the item etc and then also defining the mapping of item name to category.
Worked well enough but I wasn't willing to pay for this after all the free trials and such if it couldn't be monetized by releasing the app. And since the fingerprinting is pretty cumbersome and error prone (Costco doesn't even have a global standard for their receipt format, never mind different gas stations etc) and I neither have the expertise nor the necessary training data for a machine learning approach.
All the data captured over the years omits very fine detail for the most part; things like weights and volumes and stuff. The data is fairly messy but there should have some signal within it however. This motivates me to revisit what sort of receipt OCR stuff is out there, it's been a TODO for a long time. I'm looking for mainly partially automated stuff, fully automate the frequent vendors, manual entry the rest: https://gist.github.com/jxramos/db8195d1972b6d7f7eb0e97be51d...