Hello. I have experience with image processing and ocr.
I usually use AWS to build the models and make the API because it's easier and faster, but this depend of the task.
Another way is to build a API in python flask and make the deploy in a machine, but for this is necessary more work that use AWS, but i need more details of the task if will be need some more automation beyond the API and image processing.
About the OCR i have two option is the Tesseract(Open sorce.) and/or AWS textract both not have 100% of accuracy but are some of the best options currently.