I think this might be a way that such a project could be accelerated. Machine learning is good, but math is extremely ambiguous. Symbols are re-used and their orientation relative to other symbols changes their meaning. Some problems want you to simplify, others want you to expand something. This project at least removes symbol ambiguity in written work.

Even if you could make a computer understand how to solve a bunch of problems, I'm not certain that blind machine learning would produce a good feedback system for getting students back on course when they misunderstand something. They could probably find some likely candidates where students may have made a mistake, but I don't think anyone has tech anywhere near this right now.