Old school AI isn't dead: How we achieved a 12× speedup on an NP hard problem