Using Rasa just as a dialogsystem to communicate with backendsystem


(Moritz Mezger) #1

Hey :slight_smile:

Im building a system, that functions as a link between the user and a Conversational Case Based Reasoning System. Currently im simply using Rasa NLP to extract Intent and Entities and a small python programm forwards it to the CBR System. Now the CBR system decides on which action to take next, and creates an NLG prompt. This works quite alright, but it restricts the dialog. I want to include different options, give the user to ask questions about the situation, controll the maschine he/she is working on. To achieve this I want to include Rasa Core to handle the dialog, query databases, answer simple questions, but also (if needed) feed information to the CBR-System.

My Problem is: If the CBR-System is presented with Information, not Rasa Core, but the CBR-System needs to decide which action to execute next. So in my mind the CBR system would need to update the tracker and Rasa Core would have to wait until the backendsystem has done this? Or is there another way?

  1. Is this possible?
  2. How is it possible?
  3. How would I write the stories?
  4. How would I let the backendsystem update the tracker.

Thanks a lot! I have like a zillion other questions, (how entities are stored, what the difference between slots and entities is…), but having just a concept of how to achieve my goal and getting answers to the 4 questions would be great.

Mo


(Juste) #2

Hello @Moji. You can run Rasa Core as an http server and simply get the predictions made by the dialogue model HTTP API. The api has the endpoints for appeding new actions, messages to the tracker so you should be able to ‘manually’ update the tracker. Can you tell me more why you need CBR system on top or Rasa?