Something that I've daydreamed of in the past is making a new constructive solid geometry "kernel" around which a CAD application could be wrapped based on projective geometric algebra (PGA).

The algorithms it enables are fundamentally more capable and robust than traditional kernels based on linear algebra (vectors and matrices). You can do really fancy things like interpolating in space and time robustly, find extrema in high-dimensional phase spaces, etc...

This could potentially allow straightforward and robust solvers for kinematics, optimal shape finding, etc...

Every few decades there's a "step change" where some new algorithm or programming paradigm sweeps away the old approach because suddenly a hobbyist can do the same thing solo that took dozens of developers a decade in the past. I suspect (but cannot prove) that PGA is one of those things.