Implement hybrid response model in Rasa

Hi, I am trying to implement a Wizard-of-Oz test for the close domain chatbot I am developing using Rasa. I wanted to know is there any way we can modify/change the response generated by the bot. So What I want to do is to use the responses (if any) generated by the Rasa bot as suggestions to the human expert user (Wizard as it is known) and then the expert users can change/edit those responses to improve them before sending to the end-user. Hence I wanted to know is it possible to implement a kind of a hybrid response model where responses generated by the Rasa bot can be Changed/modified by human experts?

Hello and welcome to the community!

Our team has created a Wizard-of-Oz FastAPI server, to be used as middleware between the end-user and the rasa chatbot. It’s currently in the review stage, to be merged into our open source chatbot named Theano. The Merge request is here. There have been some serious limitations to our approach, will share in a later response.

We were thinking about using custom channels as a solution, but we weren’t quite sure about it then as channels are typically not made for such use.

Hello, Thank you for the reply, I will surely check the same. I was also thinking of using NLG Servers in Rasa to do so. Please correct me if I am wrong, but I was thinking to create a web service that I will configure as an external HTTP service in the NLG server, where Rasa can perform NLU and send the identified response to it. Using this I can show the original response identified by the Rasa bot to the expert user and then allow him to change/update it. Then send this updated response back to Rasa Bot. So am I right here? can I utilize NLG severs in rasa to do so? basically what is explained here - NLG Servers Thank you in advance.