Cool project! Have you seen any interesting correlations between languages, paradigms and the accuracy of your suggestions?