I have written / am writing my own budget app. Mostly cos the idea of sending some random company my bank login details just to get back a pie chart seems insane.
Anyway, I am using regexs and globbing to match transactions - but it seems there are bigger/better ideas - but I don't really know what area of AI might help?
The only way I can imagine getting a corpus larger than my own transactions is to have some communal data dump.