Hi @surajsonee. First of all, great job on describing the situation and providing the files - it made it very easy to investigate and replicate the project 
I think the main reason why your bot behaves in a bit unpredictable way is because you don’t have any stories which include cypha_app intent apart from the that short one called “cypha app”. By default, Rasa augments your training data by putting together the stories you have in your training data file, that’s why your assistant didn’t completely fail, but you shouldn’t rely on data augmentation to teach your assistant to handle important dialogue turns. I would suggest adding at least one longer training story which would include cypha_app intent. For example:
* greet
- utter_greet
* cypha_app
- action_recording
...
Another thing - in the conversation you had with your assistant you said “Ok, great”. This message will most likely be classified as “mood_great” since you don’t have any other intent to capture thank you’s and similar inputs. This very likely also messes up the predictions as well. I would suggest creating an intent “thank you” for that or something similar and adjust the stories accordingly.