What is the best structure (Accuracy of the text extracted) for building an OCR? ATTENTION, CRNNN, DRAM,RAM, CTC based

If I want to make a new OCR for extracting text from textbooks, specially maths and chemistry, what should be the structure for the OCR? THERE ARE LOT OF TUTORIALS around the internet but no one addresses extracting texts from simple images. Either they try to extract the text of book using Pytesseract or they make a new OCR on the Number plates data.

Just like tensorflow has a n attention based OCR for the same purpose. Can I use this one to extract long texts like Tesseract does? If not, please suggest an architecture for extracting the texts from books images.