Good FOSS 3+ axis CAM software, preferably OS-agnostic. There's a quiet FOSS revolution happening in machine tool motion control: several usable CAD packages (FreeCAD, LibreCAD, SolveSpace, to name a few), a powerful motion controller (LinuxCNC), even a few open embedded servo drives. But flexible, powerful CAM to glue these parts together seems to be the type of tough niche problem, requiring a lot of domain knowled…
I think you could turn that into a billion dollar business. Start a consultative CAM company which used a commercial CAM software. Work closely with businesses to perform their CAM work for them. You'd need quick turn around and likely to do a lot of traveling. You'd need to hire some top notch NC programmers. Simultaneously hire some top notch software developers with either CAM or 3D graphics backgrounds. Initially have the NC programmers do all the work manually, then have the developers automate away processes so that the CAM work gradually becomes easier. The final goal would be to have software which is capable of taking data from a library of machine/control definitions you've built up doing work and using it to go straight from a CAD file to NC code; skipping the entire CAM process. In the end I think you'd have a cloud service which users uploaded CAD files to and downloaded NC files from. You could integrate this a step further by monitoring machine performance and integrating it with file generation. At that point you're in the "How could we apply machine learning to this?" territory. The neat thing is that you'd basically have the same upload/download workflow whether it was people or software doing the actual NC code generation, and that 3rd party NC code generation is a service which companies already pay for from small contractors. The biggest challenge would be liability in running the NC code. You'd need to work out all the contracts and having knowledgeable NC programmers would be key. Open source CAM software could be a side effect of this, something developed along the way to cut on scaling costs; if that makes sense. Though, just eliminating the manual CAM process would be the final goal, so I probably wouldn't invest too heavily into that. There's a million things which could be branched out into once integrated this heavily in the manufacturing process. You'd probably need about the standard YC handout to make the initial website and prove you could get orders to do the CAM work (this could potentially pivot into cloud based manufacturing management or services trade). You'd then need a couple million to actually hire NC programmers and computer programmers and deal with business costs. If you really wanted to do it right, add a couple million on to that for in house machines. At that point you're going to make back the investment money as the Uber of CAM services, and then, if you can pull off the engineering, you automate away the CAM process all together and completely change how subtractive manufacturing is done world wide.