Ooh exciting! The main reason why I've needed to leverage the GCP vision API is due to orientation limitations on local OCR.

I'll test and migrate to this soon depending on accuracy. Great job so far.