Comment by Oras

OCR has been solved long time ago with vision models. Solutions are consistent, reliable, and stable. What is the point of reinventing the wheel?

I would definitely understand post processing, like extracting data, answering question .. etc, but why re-doing the OCR engine itself?