I've seen a lot of uses for SAT solvers, but what do you use them for in data science? I can't find many references to people using them in that context.

Root causing from symptoms is one case where SAT or their ML analogue -- graphical models are quite useful.