Since you cannot really beat Google Vision's OCR for obvious reasons, especially for handwritten text, I guess would still be really helpful to implement the current parsing logic that you have developed but on the .json that Google Vision produces so you will combine the state of the art OCR with state of the art Parsing :)