←

Back

Overview

Problem Space

The Idea

Solution

The Experience

Outcomes

Reflections

ZEP Quiz

NAVER Z CORP

Making it safe to be wrong

Making it safe to be wrong

Making it safe to be wrong

Designed and shipped ZEP Quiz's first AI learning assistant end to end – six AI pet companions that build rapport with students during live quizzes, then review each student's mistakes in a private 1:1 chat, while giving teachers a window into the misunderstandings their students never say out loud. The work shipped on a platform used by 80% of Korean K-12 schools.

Designed and shipped ZEP Quiz’s first AI learning assistant end to end – six AI pet companions that build rapport with students during live quizzes, then review each student’s mistakes in a private 1:1 chat, while giving teachers a window into the misunderstandings their students never say out loud. The work shipped on a platform used by 80% of Korean K-12 schools.

Role

Task Owner
Lead Product Designer

Timeline

7 Months

(Feb – Aug 2025)

Team

NAVER Z Corp.

ZEP Quiz AI TF

Context

Context

ZEP Quiz is in 70% of Korean K-12 schools. When the quiz ended, the learning stopped.

ZEP Quiz is in 70% of Korean K-12 schools. When the quiz ended, the learning stopped.

Problem

Problem

Mistakes went unreviewed. Students had no safe, personal way to ask about them.

Mistakes went unreviewed. Students had no safe, personal way to ask about them.

Solution

Solution

I designed and shipped six AI pet companions: they build rapport during live quizzes, then review mistakes in a private 1:1 chat. Teachers get an insight view.

I designed and shipped six AI pet companions: they build rapport during live quizzes, then review mistakes in a private 1:1 chat. Teachers get an insight view.

Outcome

Outcome

After user testing, it shipped as a Premium plan feature. Within 2 months, paid-user usage reached 82%, reuse reached 68%, and students sent 9.3 chat messages on average.

After user testing, it shipped as a Premium plan feature. Within 2 months, paid-user usage reached 82%, reuse reached 68%, and students sent 9.3 chat messages on average.

Paid-user usage

82%

82%

82%

within 2 months of launch

Avg. chat messages

9.3

9.3

9.3

sent per student per quiz

Reuse rate

68%

68%

68%

teachers set the feature again.

Product Background

Product Background

What is ZEP Quiz?

What is ZEP Quiz?

ZEP Quiz is NAVER Z's gamified quiz platform, used in 70% of Korean K-12 schools.

ZEP Quiz is NAVER Z’s gamified quiz platform, used in 70% of Korean K-12 schools.

37M+

37M+

37M+

cumulative quiz plays

cumulative quiz plays

2M+

2M+

2M+

monthly active users worldwide

monthly active users worldwide

Problem Space

What happens after the quiz ends?

What happens after the quiz ends?

What happens after the quiz ends?

Students solve quizzes in real time while teachers watch class-wide progress. The flow had three stages – Content Learning, Quiz Session, Result Check. One was missing: Review & Feedback. Once a quiz ended, mistakes went unreviewed and gaps widened.

Students solve quizzes in real time while teachers watch class-wide progress. The flow had three stages – Content Learning, Quiz Session, Result Check. One was missing: Review & Feedback. Once a quiz ended, mistakes went unreviewed and gaps widened.

Content Learning

Quiz Session

Result Check

Review & Feedback

Field research

I went into real classrooms – observations, student interviews, teacher focus groups – to see what actually happens after the quiz ends.

I went into real classrooms – observations, student interviews, teacher focus groups – to see what actually happens after the quiz ends.

In their words – verbatim from field research

Asking means exposure.

Verbatim quote from a student interview – what they said about asking questions after a quiz. Korean original is fine.

Review time doesn't exist.

Verbatim quote from a teacher interview – what they said about review time or individual follow-up.

Thirty classmates are watching.

Verbatim quote that captures the fear of judgment in front of 30+ classmates.

“I can't get to every student.”

Teachers face 30+ students at once and cannot address individual misunderstandings in real time.

“I just move on.”

After quizzes, most students never ask about their incorrect answers.

“I have to teach to one level.”

Classrooms hold wide differences in understanding, but the lesson can't.

Students weren't unmotivated. They were unsafe.

Students weren’t unmotivated. They were unsafe.

Students weren’t unmotivated. They were unsafe.

I initially assumed this was a motivation problem. Field research showed the opposite: students do have questions – fear of judgment in front of 30+ classmates, and no structured moment to reflect, keeps them silent.

I initially assumed this was a motivation problem. Field research showed the opposite: students do have questions – fear of judgment in front of 30+ classmates, and no structured moment to reflect, keeps them silent.

The Idea

What if review felt like talking to a friend?

What if review felt like talking to a friend?

What if review felt like talking to a friend?

I reframed review as a relationship problem: not a tool students use, but a companion they trust with their mistakes.

I reframed review as a relationship problem: not a tool students use, but a companion they trust with their mistakes.

How Might We

Make review feel safe, personal, and naturally embedded into the quiz flow?

Make review feel safe, personal, and naturally embedded into the quiz flow?

Make review feel safe, personal, and naturally embedded into the quiz flow?

Three pillars of the companion concept

01

01

Rapport before review

Students open up about confusion only after they trust the assistant. Trust is built during the quiz, not after it.

02

02

A friend, not a teacher

Asking for help should feel like talking to a companion, not raising a hand in front of 30 classmates.

03

03

Reflection embedded in the flow

Review can't be a separate destination students must choose to visit. It has to live where the quiz already is.

Solution

AI Pet friends for personalized review and support

AI Pet friends for personalized review and support

AI Pet friends for personalized review and support

A companion that hands out generic answers breaks the spell. A wrong answer encodes how the student thinks – so I designed the interpretation layer myself, keeping each companion's personality alive on real student work.

A companion that hands out generic answers breaks the spell. A wrong answer encodes how the student thinks – so I designed the interpretation layer myself, keeping each companion’s personality alive on real student work.

The Experience

Outcome Brief

AI Pet Companions

Six companions who cheer students through the live quiz and earn their trust.

Student Review Chat

A private 1:1 space where students revisit their own mistakes.

Teacher Insight View

A window into the misunderstandings a class never says out loud.

Part 1

AI Pet Companions

AI Pet Companions

AI Pet Companions

Six AI pet companions build rapport during live quizzes – cheering, reacting, celebrating – so asking for help afterward feels like talking to a friend, not raising a hand in front of 30 classmates.

Six AI pet companions build rapport during live quizzes – cheering, reacting, celebrating – so asking for help afterward feels like talking to a friend, not raising a hand in front of 30 classmates.

Personalites

AI review chat, in action

Drop a short looping screen recording here – phone ratio (9:16), autoplay + muted + loop. Show the pet responding to a wrong answer.

Part 2

Student Review Chat

Student Review Chat

Student Review Chat

A private 1:1 chat where students revisit their own missed questions – no audience, no judgment, no separate destination.

A private 1:1 chat where students revisit their own missed questions – no audience, no judgment, no separate destination.

User Goals & Features

Private 1:1 Space

Students revisit missed questions in a private chat, anytime – no audience, no judgment.

Personal Explanations

Feedback is generated from each student's own written reasoning, not a generic answer key.

Embedded in the Flow

Review lives where the quiz already is, not in a separate destination students must choose to visit.

My Score screen

AI chat thread

Part 3

Teacher Insight View

Teacher Insight View

Teacher Insight View

Teachers see each student's learning gaps and class-wide misunderstanding patterns through chat details – and use them to adjust the next lesson.

Teachers see each student’s learning gaps and class-wide misunderstanding patterns through chat details – and use them to adjust the next lesson.

User Goals & Features

Individual Gaps

Teachers see each student's learning gaps and misunderstanding patterns through chat details.

Class-Wide Patterns

When multiple students share the same misconception, the pattern surfaces across the full chat history.

Next-Lesson Adjustment

Teachers use what the class could not say out loud to adjust the next lesson.

Teacher LMS

with chat history list

Image needed

In the wild – a teacher running the review flow in class

Outcomes

From user test to paid adoption.

From user test to paid adoption.

From user test to paid adoption.

After user testing, the assistant shipped as part of the ZEP Quiz Premium plan.

82%

82%

82%

Paid-user usage

Within 2 months of launch.

68%

68%

68%

Reuse rate

Used the feature again.

9.3

9.3

9.3

Avg. chat messages

Sent per student on the page where the feature was used.

How I validated it

Prompt evals in LangSmith

50+ prompt versions scored against real student answers – quality measured, not assumed.

Live classroom pilot

Students used the assistant in real quizzes during the pilot – behavior observed in the classroom, not in a lab.

Teacher feedback

Pilot teachers reviewed the insight view and shared what changed in their classrooms.

What we observed

Active engagement

Students who normally stayed passive became active participants, voluntarily opening the review chat after quizzes.

Deep immersion

High focus maintained through personalized AI interaction.

Psychological safety

The private, non-judgmental chat lowered the emotional barrier to asking questions.

of students opened the review chat on their own – no prompt, no assignment.

of students opened the review chat on their own – no prompt, no assignment.

of students opened the review chat on their own – no prompt, no assignment.

[Number needed]


Piloted in [classrooms / students / schools / duration].

Reflections

What I learned.

What I learned.

What I learned.

1

1

1

High student engagement vs. fear of replacement.

Students engaged actively, yet some teachers hesitated, perceiving AI as a threat to their role rather than a support tool. Adoption depends on teacher trust as much as student delight.

Students engaged actively, yet some teachers hesitated, perceiving AI as a threat to their role rather than a support tool. Adoption depends on teacher trust as much as student delight.

2

2

2

AI can lower the emotional cost of asking questions.

A private, non-judgmental space mattered as much as the correctness of the explanations.

→

→

→

Next Project

ZEP Quiz Main Redesign

ZEP Quiz Main Redesign

ZEP Quiz Main Redesign

Case study · NAVER Z · 2026

2026 Cora Lee

cora

Work

Contact

About

Resume