Pon IT – Chatbot Rasa to Dialogflow

The goal of this project was to transfer Pon IT’s chatbot from Rasa to Dialogflow. Dialogflow offers build-in NLP and has an integration with Hangouts Chat (and other message services). Dialogflow offers the possibility to add intents (what does the user intent to say) in order to come up with a response. Dialogflow recognizes the intent based on training phrases. Responses can either be text, buttons or some other custom payload. Dialogflow accepts bulk uploads of training phrases, intents and responses in a particular json format, which speeds up the process significantly.

Through the use of a fulfillment API we can orchestrate the response of atypical intents. We can enrich the response in the fulfillment API with other data sources, ServiceNow information for example, and afterwards send the desired response to Dialogflow.

Interactions with the chatbot are registered in a standalone RDS database. Based on this, a Tableau dashboard provides insights into the interactions with the bot. The dashboard can be used to monitor the conversations and to improve the chatbot.

Possible next steps would be to improve the quality of training phrases and responses. Additionally, it would be interesting to improve the quality of the chatbot through the introduction of Context in Dialogflow. Adding follow-up intents would be a good starting point.

  • Kasper Meijjer

    Senior Data Scientist