Writing a program that analyzes traditional Scottish music and composes new tunes in the same style. Right now I'm in the careful planning phase--reading up on LSTM recurrent neural networks and such, but I should be able to start coding shortly (at the very least I can do all the software engineering parts while I figure out what the best algorithm to use is).

I'm not deluding myself about its long-term non-profitability...I'm just doing it because I find it fascinating. At least there's no hosting cost--I already have a laptop running SBCL and Emacs, which is all I need.