I haven't tried yet, but this might be already possible by generating C code and then building a Python package from that. It would be nice to have that automated.
I think there's also a commercial potential to sell Vision Assistant as a separate product (to Python developers).
