Spike: Changes to Rasa Open Source / X needed to hit e2e ready for production goal

[TyDunn]

A definition of the goal and more context for this spike can be found here.

Primary question to explore with this spike

What would we need to adapt in Rasa Open Source to make workflows work and in Rasa Enterprise to ensure it does not break?

Secondary questions to motivate exploration

Rasa Open Source

• Would anything about training need to change?
• Would anything about testing need to change?
• Do we need to rethink test conversations which requite an intent assigned to each user message? Would anything else about evaluation need to change?
• Would we like to make intent assignment optional in interactive learning? Would anything else in interactive learning need to change?
• Would anything about rasa visualize need to change?
• Would anything about data validate need to change?
• Would anything about rasa shell —debug need to change?
• Do we need to adjust or construct to support user key inside it? Would anything else with training data need to change?
• Would anything with forms need to change?
• Would anything with rules need to change?
• Would anything with the domain need to change?
• Would anything with fallback (including two stage) need to change?
• Would anything with the response selector need to change?
• What would we need to do to get the the /restart command to work?
• What would we need to do to get the /<intent_name> commands to work?
• What needs to change for incremental training to work with E2E?

Rasa X

• Would anything with training need to change?
• Does story persistence / loading from DB + visualising work? Would anything else with story visualisation to change? Would anything with else how stories are dumped to training data need to change?
• Would anything intent insights need to change?
• Would anything with the training data screens need to change?
• Would anything with the domain need to change?
• Would anything with IVC need to change?
• In Rasa X, there won't be any indication whether the next bot action was taken based on the intent detected or the text of the user message. Additionally, the presence of intent next to each user message in the conversation screen will be confusing to users. This will hamper the user experience of the feature. Is this breaking any of these workflows?

Definition of done

• Document created that shares a summary of the spike exploration and findings
• Recommended next steps agreed on with e2e validation project (@dakshvar22 @JEM-Mosig @TyDunn)