Hi, it should be definitely possible. We primarily used RASA (Open Source/community version) to reduce the training costs.

The code already has a Watson mapping. You can similarly use LEX.

The code is available at:

https://github.com/debmalyabiswas1/ReinforcementLearning_Chatbots

Dependencies:

- rasa with tf pipeline [not needed if using watson ]

- watson_developer_cloud [not needed if using rasa ]