ChatBot that uses Reinforcement Learning with Python (TF/keras)

I need to have Chat Bot that uses Reinforcement Learning for educational purposes. The main idea is too use Seq2Seq model and then Reinforcement Learning method/methods to make it better. It should not be Goal Oriented Chatbot, rather general topic, dialogue generation with RL (ex. https://arxiv.org/abs/1606.01541). I need comments to RL part, so i will not have to reverse the source code to make changes. The Seq2Seq model is not my goal, it can be everything. Like pretrained model from the github or other base. The main goal is clear RL part, which i can understand and read clearly, and prove that RL made or did not make the chatBot better. If there are others ideas, maybe simpler for chatBot with RL, i am open for suggestions. It does not have to be very well trained, at least 2-4h on personal computer, just to see the diffrence. I tried The Cornell Movie-Dialogs Corpus as dataset, when i did my own seq2seq, so i propose this one, but feel free to choose.