I can recommend a big stupid project where it's not a big deal if you fail.
Until recently, for me that would've meant "run Windows programs on Android," mashing together Wine and e.g. Unicorn Engine... but now there's like six different "userland emulators" vying for preference. FEX-emu, Box86, others with even sillier names.
I did manage a bespoke 8-bit FPS. That only took about two months. And then I've been idly tweaking it over the last two years. Splitscreen multiplayer, as a joke, was maybe not the best idea.
What'd take both time and space is some extremely low-end VR. I am convinced that Quest-ish headsets could cost, like, fifty bucks. The big players keep iterating clever hacks from a decade ago. Solving those problem, instead of avoiding those problems. Light should be collimated by default, which means a point source, which is any single LED. Rendering has to be detached from software performance, which means projecting the nearby world to e.g. floating dots instead of directly making a flat image. Inside-out tracking is at least the right idea, but it doesn't have to be especially good in order to ground an inertial estimate.