We have some rough ideas to support multiple "backends" depending on the input, so that could definitely happen. That being said, tools like LaTeX-OCR [1] already claim to (at least partially) support this.

[1] https://github.com/lukas-blecher/LaTeX-OCR