Reports of building code violations range from the silly to the disgusting, and with hundreds of different building codes, how can one possibly analyze trends effectively? Grouping the codes into major categories will help, but it will take the power of natural-language processing (NLP) to successfully distinguish between a busted sewer line and a busted window. We'll use a case study of Metro Nashville code violations analysis to get a high-level understanding of concepts like word vectors, Naive Bayes classifiers, and model training, among other things. We'll also talk about model fitting difficulties, and how to decide when a fit is "good enough".