Sila, your virtual closet powered by AI
Most people have full closets but feel like they have nothing to wear. Sila acts as a co-stylist to help users create new outfits from the clothes they already own.
Year
2025
Timeline
4 weeks
Role
Product designer
Team
Solo project
Overview
Sila is a virtual wardrobe app that helps women maximize their closets. The goal was to turn closet overwhelm into outfit confidence, proving that great style comes from smart styling, not more shopping.
As the product designer, I led user research, concept development, and prototyping to solve common frustrations like decision fatigue, forgotten items, and time-consuming onboarding.
Discovery
Competitive analysis
I tested five leading virtual closet apps — ACloset, Fits, Indyx, SimpleCloset, and Whering — to evaluate usability and AI styling.
Manual onboarding was a major pain point for me, taking over 20 hours to upload 100 items. Without automation for new purchases, my closet quickly became outdated, and some apps charged over $100 annually for AI outfit suggestions that I would never wear.
User research
To understand real-life wardrobe challenges, I interviewed women ages 24–44 in major U.S. cities with active social lives and demanding professional schedules. These conversations uncovered unmet needs and behaviors that went beyond what I observed in the competitive analysis.
Defining the problem
Research analysis
Affinity mapping revealed that many women lose track of what they own, leading to underused wardrobes and unnecessary spending. While they often turn to social media for outfit inspiration, they struggle to recreate those looks using their existing clothes.
Problem framing
I translated research insights into two focused How Might We statements to guide design:
How might we help women quickly upload their wardrobe without manual uploads?
How might we generate outfits that women aspire to wear using clothes they already own?
Solution ideation
Brainstorming
I used the Crazy 8s method to rapidly sketch solution ideas for closet visibility and outfit creation.
Prioritization
Next, I created Impact/Effort matrices to identify MVP opportunities.
Gmail Sync emerged as a quick win, automating closet onboarding with minimal user effort. Style Match was a larger initiative, designed to generate outfits from user-uploaded inspiration.
Solution 1: Gmail Sync
Gmail Sync cuts onboarding from hours to minutes by automatically uploading past and future purchases — a capability no other virtual closet offers.
Sila scans fashion receipts, extracts product details, and categorizes items directly into the user’s digital closet. Gmail was chosen for its global reach and strong adoption among Sila’s target audience.
Sketching key moments
Before moving into Figma, I storyboarded the Gmail Sync journey, noting emotions below each step.
Early interaction ideas included a notification prompt so my persona, Ari, could return to what she was doing while the closet populated, and a swipe gesture to keep or delete items once they were imported from email receipts.
Early feedback
I tested low-fidelity Figma mockups with six survey respondents to validate the interaction design. Based on their feedback, I refined the onboarding flow, improved data transparency, and designed alternative user paths.
Design iterations
Addressing privacy concerns
Nearly every participant was uneasy about giving Sila access to Gmail due to privacy concerns or uncertainty about the feature.
To reassure users at the decision point, I designed a “How does it work?” tooltip with progressive disclosure. This opens a bottom sheet explaining the process and emphasizing that only receipts from trusted retailers are scanned, reducing perceived risk without interrupting task flow.
Designing with system limits in mind
In my initial mockups, I used a placeholder of seven minutes to upload 123 items. When a participant questioned this, I realized I didn’t know how accurate it was. To ensure the design reflected actual backend capabilities, I mapped the Gmail Sync system architecture to understand its performance limits.
Results:
Added backend logic to automatically filter returns, removing the swipe interaction and eliminating the need for users to review hundreds of items.
Parallel processing enables 100 items to upload in 2–3 minutes, allowing me to set more accurate expectations during onboarding and loading states.
Classifying images during upload allows Style Match to find closet matches faster and with greater accuracy.
Providing another way to get started
Testing showed some users thought Gmail was required, creating a barrier to entry.
To show Sila’s value beyond sync, I designed a flow that lets anyone start with popular basics, which could later be personalized by location and season to show items the user is more likely to own.
Final designs
The final Gmail Sync flow blends AI and proven technology for speed, reliability, and cost efficiency. AI handles high-value tasks like extracting receipt data, validating images, and analyzing fashion attributes, while other processes use stable, traditional tech.
Mapping the system architecture showed onboarding can be fully automated — cutting upload time for 100 items from 20 hours to 2 minutes with 94% accuracy. Once launched, Sila will be the only virtual closet app that automatically builds and maintains a user’s wardrobe.
Solution 2: Style Match
Style Match turns inspiration into real outfits using clothes the user already owns. After uploading a photo of an outfit they admire or are currently wearing, Sila’s AI analyzes the style, color, and silhouette to suggest similar combinations from their closet.
Sila does not use AI to guess what someone wants to wear—it uses direct user input to create outfits users will love to wear.
User testing
I conducted task-based testing with eight participants. Key friction points included the immediate need for inspiration images, lack of AI transparency, and limited user control over generated outfits.
Design iterations
Providing inspiration to encourage action
Testers felt pressured to have an inspiration photo ready before using Style Match. I redesigned the flow with a bottom sheet that introduces the feature and offers multiple ways to start—removing that pressure and making it easier to begin creating outfits.
Enhancing visual affordance
Testers didn’t realize matched items were tappable. I updated the closet item component to signal interactivity without distracting from the outfit.
Improving AI match transparency
Testers were skeptical of how match scores were calculated, which made them question the effectiveness of Style Match. To build trust, I added a clear explanation of how scores are generated. I also introduced an “Other Matches” option, giving users the freedom to adjust and make it their own.
Balancing automation with control
To give users more control, I designed a tap-to-reveal menu for swapping items, shuffling looks, and customizing outfits. I prioritized progressive disclosure to keep the interface clean while making advanced controls easily accessible.
Supporting creative exploration
Some users expressed the desire to style looks freely without uploading photos. Freestyle gives them a simple, flexible way to explore combinations from their closet.
Final designs
The final Style Match experience solves a real problem: people can dress the way they aspire to using clothes they already have. Users stay in control throughout by choosing their inspiration, experimenting with results, and making the final decision about their look.
The result? Less shopping, less stress, and more confidence in everyday style decisions.
Learnings
Throughout the design process, I uncovered insights that shaped product decisions and refined my approach to AI-powered UX.
System insight is crucial to UX
It’s okay to make assumptions early, but I learned how critical mapping workflows is to the user experience. Grounding my designs in technical reality helped Sila reduce unnecessary user effort and optimize future interactions like Style Match.
Transparency builds trust in AI
Match score breakdowns helped users understand how outfits were generated, turning doubt into confidence. This taught me to prioritize transparency from the start so trust is built into the experience, not added later.
Anticipate ethical use cases early
I overlooked the risk of using Pinterest images, which could require major product or strategy changes. This raises an important question for Sila — how should the product handle AI styling or recommendations when the source images may not belong to the user?
Next steps
I’m currently vibe coding Sila into a beta app for testing with friends and family. If you’d like to join the beta list, email me at Alea.nalesnik@gmail.com — I’d love your feedback.