How to train tesseract to recognize more complex characters?

Question

C++ / C#

Guys, can anyone come across the task of learning tesseract? I would really like to add such functionality to my program, but I can’t figure it out. As I imagine it: tesseract receives a picture of a word as input, it recognizes it, I correct it if necessary, and thus it updates its tessdata dictionaries and in the future it will correctly recognize this word! At the moment, its recognition itself leaves much to be desired ... I slipped the text binarized both by paragraphs and by lines and by words and even by characters, by the way, the characters turned out to be the most terrible recognition, for some reason ... And now I have my last hope for training
I will be glad to any advice!