I was asked about the way I would design the "people you may know" application of LinkedIn and about the way I would group "similar" people and jobs with people etc. More or less it was a discussion about collaborative filtering , classification and prediction models (k-means clustering, bayesian model, decision trees ).