OCR plus some LLM prompts would be a more general solution than regex, if it works. But if Veryfi works than I wouldn't bother. (LLMs are probably what they use anyways...)
Sincerely, Timothy Jesionowski On Sat, Jun 21, 2025, 1:25 PM TRS-80 <[email protected]> wrote: > Martin Blais <[email protected]> writes: > > > These days for OCR I think you can just download a free vision model > > from Hugginface and run it locally and it would work. > > I remember doing that in the recent past. > > I imagine something like that just returns a big blob of plain text, > amirite? Or can you have it return the data to you in a more structured > format, with specific fields (key-value pairs) more relevant to a > receipt? > > With this Veryfi API, they are returning JSON with very specific fields > (e.g., card number, date, payee, tax, tip, total, even individual > receipt lines (including UPCs when avalable), etc.) and it seems to be > very accurate so far for me. > > When I played with other general OCR tools in the past, I remember the > OCR itself was only half the battle. Even if you got that accurate, you > then had to write regex trying to pull all this other specific info out > from that. > > -- > Cheers, > TRS-80 > > -- > You received this message because you are subscribed to the Google Groups > "Beancount" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion visit > https://groups.google.com/d/msgid/beancount/87bjqh7y19.fsf%40isnotmyreal.name > . > -- You received this message because you are subscribed to the Google Groups "Beancount" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/beancount/CAOVsoWRzf7FGZaE-P%3D1URKqkvQSOQVcHc%2BKHUe6OANSMfmsoVQ%40mail.gmail.com.
