I feel that the Human In The Loop should have an option for creating a new qna instance or a new intent when a user asks a novel question or obscure question.
I also feel that nlu debugging should be made simple by allowing an admin to re-assign a test phrase (question) to an answer which is appropriate and automatically giving a confidence level of 1 when the bot deduces a wrong user meaning.
Although I tried to create a python + scikit model for these, but its not gelling well with the Typescript sdk and I am hard pressed for time to study the sdk itself. But for these few features I am sure they should be fairly easy to implement.
And you may have noticed that I only post on the community forum which is because of some organisational policies in relation to GitHub (Please bear with me there).
However, I am always more than happy to interact with folks on these issues.