Cool! I was also working on a prototype that uses text-based reinforcement learning techniques. This has a different use-case, but good to read anyway.