This case study documents how I designed a fashion marketplace around real fan behavior, not new search mechanics.
TIMELINE
5 months
(2023-2024)
ROLE
Worked as a Product Designer
COLLABORATORS
Product Manager, Front/Back-end developer, AI engineer
COMPANY
Teamsparta Inc. (Edtech SMB
Achievement
Overall mobile APP screeNs
Overall WEB screeNs
When inspiration turns into manual investigation
Image source : INSTAGRAM Account (@style_Moabom)
Define problem
The desire was never the problem. The system just wasn’t built for how fashion spreads.
How might we
Asking real world again
We spoke with Gen Z shoppers. The behavior was already there.
72%
Take outfit screenshots weekly
Screenshots are how inspiration is saved
92%
Couldn’t find the exact item later
Discovery breaks after the moment passes
67%
Tried describing outfits in search
Visual intent gets lost in words
58%
Gave up mid-search
Friction dissolves purchase intent
The problem wasn’t inspiration. It was the lack of a system that could carry interest forward.
So I asked: What if discovery didn’t end at inspiration, but naturally evolved into exploration and action?
Hypothesis
Our hypothesis was simple.
If fashion discovery starts from
how people already browse celebrity outfits,
users can move from interest to action without switching tools or re-searching.
We hypothesized that organizing fashion by celebrity appearances and outfits would shorten the path from discovery to purchase.
solution walkthRough
I started by anchoring the product in an existing behavior, rather than introducing a new way to search.
Screenshot search - Pull based behavior
Screenshot based discovery
Users start from an image they already saved. Walaland detects the top and bottom automatically and matches them with real brand products.
This flow does not replace search. It supports the behavior users already have.
01
Users encounter a look on Instagram, TikTok, or in media and save it as a screenshot.
02
They upload the screenshot to Walaland. The system separates outfit components without requiring any text input.
03
Walaland returns purchasable matches from real brands and keeps the user inside the app to continue browsing.
Style-to-brand redirection - Push based behavior
Once a screenshot becomes a starting point,
users need a way to keep exploring without starting over.
Users exploring celeb lookbooks or moodboards can tap on any outfit -
and get instantly redirected to where that item (or the closest match) lives.
No more asking strangers for links and waiting. Just tap and go.
Exploring Walaland app
Users open the app and freely browse content. They see photos of outfits worn by celebrities, and can tap to get redirected to sites selling similar items.
Exploring Walaland app
Users can also swipe through short-form videos like Reels - where outfits worn by celebrities are featured and made shoppable.
Designing with AI in the loop
And behind that simplicity, 38-layer AI engine was working quietly in the background.
I designed the system to surface these signals clearly - without users needing to ask. Instead of dropdowns or filters, the UI simply showed what the AI saw : structured recommendations, top/bottom splits, and scrollable matches by type.
AI understood garment strucuture
WALA’s backend used a 38-layer deep learning model to scan a screenshot and detect visual signals:
tops vs bottoms, sleeve type, neckline, overall silhouette.
But raw detection alone wasn’t enough.
Designing with AI in the loop
After finalizaing our 1st version of product,
We tested with Gen Z K-pop fans, and listened. Closely.
These two personas revealed one clear insight: Not all users shop the same way - but both expect the product to understand their intent. Our hypothesis was simple: If we designed a visual-first experience, both vibe-driven browsers and brand-driven hunterscould find what they love - in their own way.
Persona 1
Emotion-first user
“I don’t know the brand… I just want more things like this.” Scrolls TikTok and K-drama screenshots daily, loves the feeling of an outfit, not the name. For her, discovery means vibe first, shopping later.
Titok lover
Moodboard scroller
Screenshot collector
Casual discoverer
Style-driven, not brand-driven
Persona 2
Brand precise hunter
“If I fall for something, I need to know the exact brand.”
Tracks airport fashion and styling tags, builds moodboards by idol. Wants to buy the real item - not just something close.
Style perfectionist
Airport fashion tracker
Brand ID detective
Idol stylist follower
Exact-match shopper
The result? Both types found what they were looking for, whether it was a mood or a match.
REFLECTION
What stayed with users wasn't the AI.
It was how natrual the experience felt.
Looking back, this project taught me,
