A couple years back there were some traders talking about trying to implement a really fast version of K-means for clustering stocks that are transiently correlated on some sub-reddit. I wonder if this guy is the one who started that thread.