The First TextWorld Challenge was a global machine learning competition organized by Microsoft Research. The competition had the goal of building agents capable of playing text-based games. The agents could not be a collection of hard-coded rules, they had to learn how to generate the text commands from playing games from a large collection provided.
Our team won the competition with a raw score of 91.9% (70.8% with handicaps). Microsoft published an article with details of the competition and our solution.