Generative AI added to app chatbot

The UniWellbeing system offers students a variety resources including podcasts, guidance and habit forming tools designed to improve mental and physical health. Generative AI deployed to improve the user in-app chatbot experience.

The UniWellbeing chatbot is configured as a series of prompts and response options eventually directing users to one of the app’s many resources. The client asked Exploding Phone to integrate natural language based AI into the chatbot to facilitate free text responses from users.

AExploding Phone had previously implemented a web-based chatbot admin portal to define and manage prompts and fixed sets of responses which then direct the user towards the most appropriate in-app resource. The system was updated to enable users to input free text responses to prompts. The anonymised text response and contextual information is then sent to the generative AI server which returns both a text response and identifies what it decides is the best match to the set of next-steps as defined within the chatbot admin system. The list of allowed responses also include handling interactions that are outside the scope of the UniWellbeing system. Based on the response from generative AI server the user may arrive at a further chat prompt (free text or fixed set of responses) or be directed to a specific wellbeing resource.

The value and commercial success of UniWellbeing is predicated upon helping students access and act upon the most useful in-app resources. Adding the ability to interact with the chatbot via free text delivers a significant improvement to the user experience and hence overall utility of the system.

Advances in AI technology allow software developers to add Natural Language Processing (NLP) functionality to their systems. The incorporation of NLP within the UniWellbeing chatbot enables prompt designers to specify a richer set of possible outcomes from each chat interaction. Tests have shown that responses from the AI are highly aligned with responses where users are presented with specific response choices.