NPS(User Satisfaction)
+
84%
,
Mactching Accuracy
+
NN%
Designed the conditions for emotion-matching to work, and made the call to pivot when AI accuracy wasn't enough.
Lead UX Designer
: Leading UX Design with Branding Design
PM
1/DS
1/FE
1/BE
+Marketing
+Psychologist
23.03~23.08(5Months)
Background
Request: Build a community app where married women can form genuine connections through AI-powered emotion matching
Objective: Design the conditions for emotional expression and relationship formation, measured by diary engagement and matching quality
A founding team was building an AI-powered community app for married women's isolation, I joined as Lead UX Designer to direct the full experience.
The goal wasn't to build a communication channel. It was to design the conditions for genuine relationships to form: integrating an AI emotion-matching system into a coherent app experience that enabled women to surface their emotions for the first time.
I started with two research tracks: a survey of 90 married women and a psychiatrist interview, to define what actually creates intimacy and what genuinely supports emotional relief. Every design decision followed from there.
user interview
How Can I make less lonely through the app?
What's the most useful way to elavorate —?
We began with a simple but critical question: Why do married women feel isolated even when communication channels exist? Married women have access to messaging apps, community forums, and social networks — yet emotional sharing rarely happens. The problem wasn't the lack of space. It was the lack of conditions.
point
Defining what experiences maximize intimacy in
community-based connection
who
A total of 60 married women ranging from
their 20s to 60s
when
23.04.17 - 23.04.23
how
User Survey
Insights :
Loneliness persists not because users lack ways to talk, but because existing systems fail to match and sustain conversations at the psychological level.
Emotional distress after marriage is widespread, not marginal
+
View More
Talking is the primary coping mechanism, but not an effective one
+
View More
Users seek emotional alignment, not surface similarity
+
View More
High communication frequency does not reduce loneliness
+
View More
The core friction is meaning mismatch, not usability
+
View More
Competitor Analysis
Building Trust & Connection: A Smarter, Safer Conversation Experience
01
Women's verification process
through onboarding
Guidance on the identity of women-only communities and verification of users' identities.
02
AI Various filtering & search
features to find partners
Guidance on the identity of
women-only communities and
verification of users' identities.
03
real-time conversation
matching
UI notation for real-time conversation
matching and sorting settings.
04
Management system needed for clean conversations
Provision of reporting functions and
AI-managed clean system features.
Challenge
How might we build a matching system that understands nuanced language and infers reliable psychological signals, even with limited data?
We introduced AI to enhance the accuracy of the matching system by analyzing chat histories.
However, as an early-stage product, we faced a fundamental limitation: The lack of sufficient, high-quality data.
01
Human Language Is Ambiguous and Contextual
User inputs contained metaphors, emotional contradictions, and implicit meanings that could not be easily reduced to keywords without losing intent.
02
Manual Query Mapping Did Not Scale
To stabilize the system, we initially attempted to map user expressions into predefined categories. However, covering the full range of emotional language quickly became unmanageable and brittle.
03
Early-Stage Data Was Insufficient
With limited user volume, the AI lacked enough examples to learn reliable language–emotion mappings, amplifying prediction variance.
Hypothesis
Enhancing Matching Accuracy by Pivoting from Chat-Based Signals to
Diary-Based Expression with onboarding process
03
Design for Longitudinal Signals, Not One-Off Inputs
04
Constrain AI Interpretation with UX Guardrails
01
Stabilizing AI Matching Through Diary-Based Signals and Intentional Onboarding
02
Replace “Keywords” with “Psychological Signal Buckets”
Test & improvement
Enhance the diary experience, shortcutting check-in the mood
The matching settings feature and ideation for chat screen layouts were conducted,
and screen-specific mockups were derived for A/B testing. As a result, the current conversation
matching settings and chat interface have been finalized.
solution 01
Onboarding Process
Creating a Safe and Personalized Space for Meaningful Conversations
solution 01
AI Partner Matching
Creating a Safe and Personalized Space for Meaningful Conversations
To ensure a secure and welcoming environment for women, I designed the onboarding experience with a thoughtful verification process, confirming identity through questions tailored specifically to women's experiences.
I made connecting easier by letting AI match users based on mood and interests. To ease conversation, profile cards highlight key topics, so starting a chat feels natural—even for shy users.
solution 02
Daily emotion note
Efforeless Emotion Recording & AI Emotion Analysis
To make daily emotion tracking effortless, I replaced long journal entries with emoji stickers and a calendar history, making self-reflection quick and intuitive. To build a journaling habit, I added reminder set up while AI-powered reports with mood graphs and keyword analysis provide deeper insights without extra effort.
design system
Designing for Emotional Comfort : A Soft and Cute Theme
That Creates a Safe Space for Korean Women
A warm pink palette with rounded shapes creates psychological comfort and a cozy, empathetic feel, aligning with research showing 95% of Korean female users prefer friendly visuals and adhering to Material Design accessibility guidelines.
Achivement & Reflection
The user satisfaction rate is 84%
After conducting around 200 beta tests, we achieved our initial goal of 100% user satisfaction. The results revealed a strong user desire for more expressive communication.
Usage patterns showed frequent reliance on system emojis in chat, leading us to enhance emoji-based emotional expression. Building on this insight, we further explored the feasibility and scalability of audio messaging to support richer communication needs.