Migrating Trulet’s Chatbot, Trudi

Trulet are the Australia-based creators of Trudi, an innovative chatbot developed to improve and streamline the rental experience for both property managers and renters. Their goal is to support their customers by providing regular updates, quick replies, rental documents, and answers at their fingertips. The Trulet team approached CDI with a specific objective: to migrate Trudi from Dialogflow ES to Dialogflow CX.

As Trulet no longer believed that Dialogflow ES could meet their requirements due to the growing complexity of the dialogs they were designing, they approached CDI to help them successfully take on a migration.

This however presented the company with a series of challenges. Many of the features and capabilities present in ES were either unsupported or implemented in a significantly different manner in CX. This necessitated careful planning and adaptation to ensure a successful transition. The Trulet team, in collaboration with CDI, had to navigate these disparities and make appropriate adjustments to ensure Trudi's functionality was maintained while taking advantage of the advanced capabilities offered by Dialogflow CX.

The project began with a comprehensive assessment by CDI of the training data and documents provided. We carefully examined the disparities between Dialogflow CX and ES, and subsequently analyzed Trudi's existing dialogs in relation to these differences. This process allowed us to determine the necessary changes and devise strategies for their implementation. Collaborating closely with the Trulet developers, we delved into understanding the intricacies of the integrations, ensuring a smooth transition from Dialogflow ES to Dialogflow CX.

Furthermore, our initial review revealed potential opportunities to enhance Trudi's comprehension capabilities. Specifically, we identified issues with the utilization of entity structures and recognized room for improvement in the precision of intents. To optimize the intents, we employed Qbox, a tool that enabled us to make informed decisions based on data analysis. By redefining the intent scopes, eliminating underperforming training phrases, and achieving a better intent balance, CDI was able to improve the recognition of Trudi by more than 40%.