What could be pretty interesting, is setting up an RLHF project for translating C into idiomatic Zig code.
Code translation has been a weakness of LLMs, but given two languages which are both simple, and have substantially similar semantics, I think progress could be made on function-at-a-time translation.
Set up a website where participants can paste a C function, get a translation, then clean it up, fix bugs, use idioms correctly (like defer instead of a free at the end of the function), correct the test if necessary, and submit.
Getting a reasonable volume of data is always the hard part with LLMs, but that’s no reason not to try.