You could do this quite easily with Amazon rekognition or Google image APIs. Both I believe give you the position of text in the page. If you place 3-4 known

You could do this quite easily with Amazon rekognition or Google image APIs. Both I believe give you the position of text in the page. If you place 3-4 known words on a page you can determine what manual input fields other text relates to. I have been playing around with this for paper order forms as I sell to an industry which is sometimes somewhat backwards...