Our platform offered content and assessments—but nothing in between. There was no low-stakes, low-pressure space for students to practice applying what they’d learned. And most AI tools available at the time were focused on grading or tutoring, not interaction. We also wanted to test how a machine learning model could support student growth through adaptive, low-pressure practice.

I designed the structure and content logic of an AI-driven practice tool, focusing on how students would engage, recover, and grow in realistic roleplay scenarios. My goal was to support learning through trial and feedback, not performance or grading.

I led the experience strategy for how AI would behave during roleplay: when it would stay in character, when it would pivot to coaching, and how tone, recovery, and feedback would be handled. We explored distinct AI agents for each role to keep the experience seamless—giving students space to practice, reflect, and grow without breaking immersion. I collaborated on interface design and worked closely with product and content teams to define branching logic, feedback pacing, and scenario complexity. I also explored options for short versus extended practice arcs, and began mapping what a verbal/audio-first version of the tool could become.

We created a prototype that let learners practice skills in realistic, flexible ways—with space to reflect, retry, and grow. It offered an alternative to quiz-based learning and showed how AI could support confidence-building and agency. The next phase focused on expanding access through an audio-first experience designed for learners who prefer verbal interaction or struggle with text-heavy environments.