Reducing team friction and AI memory loss
MINIai is an app where fans chat and call with AI versions of real K-pop stars. I joined for one month to work on updates to its AI features, introducing storyboards to align the team and helping fix the AI's loss of context with a bookmark feature.
Business Impact
-67%
Flow Explanations
0 → 1
Storyboard Process
N → 0
Details Repeated to the AI
Case Studies
Gave a team a shared blueprint and gave an AI a way to remember what matters to fans.
Problem
A. Immediate redirect
Upon payment completion, users are automatically taken to the questionnaire page, making it the clear and natural next step.
B. Modal alert
If a user attempts to navigate away before finishing, a warning message appears explaining that their order can’t be shipped until the questionnaire is submitted.
Constraint
The AI could not carry the full conversation history because of the token limit, and removing the limit would have raised costs too much. With only one month on the project, I prioritized the changes with the biggest impact on team efficiency and user experience.
Solution
A. Storyboards
I created storyboards to map how features and pages connected. Developers could refer to one document instead of asking for the same flow to be explained again.
A sample page and completed storyboard
B. Bookmark: user-controlled context
I tested the AI myself before the performance meetings, held two to three times a week, and shared feedback with the team. In those meetings, I worked with the AI developers and traced the cause to the token limit. To work within that limit, we settled on a bookmark feature that lets users choose which messages the AI always remembers, whether their own or the AI’s. Only saved messages stay in context, so conversations remain natural without raising costs.
Outcome
- Flow explanations dropped from about 12 to 4 per day between the designer and two front-end developers after introducing storyboards.
- Bookmark feature shipped in a new version, addressing the context loss found in testing the AI.