Real-world OCR is much harder than demo examples

Been experimenting with OCR and AI document workflows lately. One thing that surprised me: the OCR model usually isn’t the biggest issue in production. Real users upload: blurry photos dark invoices