AI-assisted design sprint
Designing with AI: a working voice-recall prototype in 2.5 weeks
Project overview
This design sprint is organized by Yummy Labs, based on a real brief from Knowunity, Europe’s fastest-growing tech education platform, with millions of students across 15+ countries. The sprint explores how to make learning stick for middle and high school students through active recall: after revising a section, a student explains the key ideas out loud in their own words.
This is an independent project, not shipped by Knowunity.
Timeline
2.5 weeks
September 2026
Role
Product designer
Skills
Product design
AI orchestrated tooling
Design system building
Prototyping
Usability testing
Team
Solo project
Outcome
Design system site and GitHub repository. Using Figma Console MCP and Claude Code, I extended Knowunity’s existing design system and published it in Storybook. Components are documented and live in code, so they’re ready for future iterations and easy to share with teammates.
Live prototype. A working prototype with real speech recognition, ready for the next round of design, prototyping and testing.
Usability testing findings. Tested with 2 students. Key findings:
1. Partial right should give more details on what’s missing from students’ answers.
2. Pause can be helpful when something happens during the recording.
These shape what to fix or explore in the next iteration.
User problem
After quizzes, flashcards and essays, a student has seen the material but hasn’t proven they can retrieve it. The existing steps test recognition or application. None of them make a student pull a concept out of memory and say it.
How can we help the student to see if they actually understand the learning material, before they reach the exam mode. And helps them prepare before the exam.
Existing steps test recognition or application, none asks for recall
Test recognition
Missing: active recall
Hypothesis
If students explain a concept aloud after revising, and the verdict feels fair, they’ll get a clearer signal of what they actually know before exam mode. In this sprint, I tested whether students would complete the loop and whether they trusted the verdict.
Some questions I hope to answer:
Would students discover the voice recall function?
Did students complete the loop?
Key decisions
Voice recall is a toggle on the exam plan
I decided existing topic nodes would just swap to a mic icon and open into a voice loop when the toggle's on, instead of building it as a separate mode.
Create a functional recall prototype
I used the browser’s real speech recognizer, so during testing I could see how users would interact with the recall process.
Provide different ways for students to move through the recall session (Never trap the student)
Students can skip a question, tap “Can’t talk right now”, or re-record their answer.
Process
Design first, then build with Claude Code, then test
Reflection
Design the core flow before reaching for Claude Code
Sketching the main flow first kept me clear on the sprint’s goal, so I was directing the AI instead of reacting to what it produced.
When I’m unsure, ask AI to explain or push back
Asking for the reasoning behind a suggestion, or challenging it when my intuition disagreed, helped me make decisions I could defend rather than just accept.
With more time, I would:
Give the partial result more detailed feedback, so students know which part of their answer was missing.
Design the switch from speaking to typing more fully, for students who can’t talk right now.
The current Home screen has a lot of information and entry points. I would look at if there is a better way to organize information, as this could be another entry point for voice recall.
