SG wardrobe

Spend 5+ mins picking an outfit? SG Wardrobe styles the clothes you already own for any occasion, SG weather, and haze—while using your body measurements to take the guesswork out of shopping.

Comment

Inspiration

I have always struggled with deciding what to wear, often spending way too long changing in and out of outfits because nothing feels quite right. Talking to my friends, I realized this is a shared daily frustration. Yet, fashion isn't one-size-fits-all—what looks great on a model or a friend might look completely different on us because we all come in unique shapes, proportions, and sizes. Because of this, buying clothes online often leads to disappointment and tedious returns. I found myself wondering: what if we could maximize the clothes we already own, and accurately analyze how new clothes will fit our exact bodies before we ever click "buy"?

What it does

SG Wardrobe is a hyper-local AI personal stylist built to take the guesswork out of getting dressed and shopping:

  • Smart Closet & Weather-Aware Styling: Users can upload their existing clothes into a digital closet. The AI then generates complete outfit combinations—including shoes and accessories—tailored to the user's occasion and Singapore's real-time weather, humidity, rain forecast, and haze levels from the official Meteorological Service Singapore and NEA feeds (factoring in everything from a humid outdoor walk to freezing 19°C indoor aircon!).
  • Body Type & Color Analysis: By inputting body measurements and skin tone, the app identifies the user's body shape and performs a personalized color analysis to show which palettes flatter them best.
  • Tailored Recommendations: Based on the user's unique proportions, the AI recommends specific clothing cuts and silhouettes, and reviews the closet to show what stays, what to restyle and what goes.
  • Smart Shopping Link Analyzer: Users paste a link to any online clothing item, and the AI reads the product page and size chart, then cross-references the garment against the user's exact measurements to recommend their best size, check the fit at the bust, waist, hips and length, and give a wardrobe match score showing how well it suits them and pairs with what they already own. When a shop has no size chart, it says so and lowers its confidence instead of guessing.

How I built it

  • Ideation & Validation: Validated core concepts, user flows, and styling logic using Gemini.
  • Development: Built and iterated on the application using Claude Code within VS Code.
  • AI: Outfit styling, body and colour analysis, wardrobe reviews and fit predictions run on Claude Sonnet 5.5 via the Anthropic API, called from serverless functions so the API key never reaches the browser.
  • Live Data: Real-time weather and haze from Singapore's official Meteorological Service Singapore / NEA feeds on data.gov.sg.
  • Backend & Database: Integrated Supabase to handle our backend database and multi-user profile storage.
  • Version Control & Deployment: Managed the codebase via GitHub and deployed the live web application on Vercel.

Challenges I ran into

  • Virtual Try-On API Limitations: Initially, I wanted to build a visual virtual try-on feature so users could see themselves wearing the generated outfits. However, existing capabilities across standard LLMs and builders (Claude Code, Gemini, Lovable) couldn't natively render this accurately without relying on expensive, specialized third-party enterprise APIs. I made the product decision to pivot away from visual rendering and focus on deep measurement and silhouette analysis instead.
  • Taking Live AI from Localhost to Production & Managing Compute Costs: Smart Shopping and the other AI features worked on localhost (through Claude Code) but failed on the live site at first. I had to set up a production Anthropic API key as a secure server-side environment variable on Vercel, then discovered every request was still failing because the account had no credits. Once both were fixed, Smart Shopping went fully live. Running live AI analysis per request also raises real questions about token costs at scale, so I limited AI to signed-in users and set a spend limit. In a commercial rollout, high-compute features like Smart Shopping and Virtual Try-On would naturally sit inside a B2C subscription tier or a B2B brand partnership model to keep unit economics sustainable.
  • Authentication Workarounds: Standard email authentication presented unexpected friction and bugs during the build. Instead of letting login hurdles block the core experience, I engineered a lightweight custom identifier system using Initials + Last 4 Digits of Phone Number. While a full OAuth/email flow will be important for future security, this nimble workaround allowed users to immediately create persistent profiles for the demo.

Accomplishments that I am proud of

I am incredibly proud of bringing SG Wardrobe from a relatable everyday problem into a fully functional, live web application! Two milestones stand out:

  1. Successfully linking the frontend to a live Supabase backend so multiple users can log in, save their measurements, and maintain independent digital closets.
  2. Debugging and deploying the live AI features, so anyone can try outfit styling and the Smart Shopping fit and sizing analyzer on real shop links, straight from the live site!

What I learned

Building this project taught me a vital lesson in product management and software engineering: a great idea isn't always immediately scalable due to API costs and build constraints—and that's okay!

Even when a feature hits a wall, it is always worth testing, debugging, and prototyping because you never know what you can get working until you try. Ideas naturally move forward and backward during development. Embracing these changes—whether that means pivoting away from visual try-ons or debugging the live AI deployment until Smart Shopping worked—leads to a much stronger, more feasible outcome. Always experiment!

What's next for SG Wardrobe

  1. Custom In-House AI Models: Rather than relying on costly third-party APIs for outfit generation, link analysis, and virtual try-ons, I want to explore building and deploying my own lightweight, specialized AI models. This would drastically cut backend compute costs and improve long-term profitability.
  2. Launching Subscription & B2B Tiers: Introducing a freemium/subscription model for heavy AI features like unlimited Smart Shopping analyses and Virtual Try-On, or partnering directly with local fashion e-commerce brands as a sizing plugin.
  3. Upgrading Authentication: Transitioning from our hackathon-friendly Initials + 4-Digit login to a fully secure OAuth/email authentication flow.

Built With

Share this project:

Updates

Submission history