The Conversations feature helps language learners build fluency through scenario-based roleplays with an AI partner. This case study explores how real user behavior challenged our assumptions and reshaped how I design for unpredictability.
Conversations is a feature designed to help language learners build fluency through scenario-based, multi-turn roleplays with an AI partner. Whether it’s ordering food or acing a job interview, the goal was to make speaking practice feel approachable and supportive while still offering actionable insights that help users improve.
In this case study, I’ll focus entirely on the moments that taught me the most: the ones where real user behavior challenged our assumptions and pushed the design to reflect more human and real-world needs. This project stretched the way I think about designing for unpredictable systems, and reminded me that designing AI interactions do not follow a script.
Why did we build Conversations?
Research revealed that users wanted a judgement-free space to practice multi-turn speaking.
Fluency was cited as one of the biggest pain points, along with pronunciation (an opportunity we covered with the Speaking practice feature).
We aimed to make the roleplay feel natural, helpful, and varied based on real-world contexts.