StyleMirror

## Inspiration

Shopping for jewellery and makeup in India is deeply personal — yet most apps treat every face the same. A wedding guest in Chennai and a festival-goer in Jaipur wear completely different palettes, and no generic "try this product" carousel accounts for that. We wanted to build the stylist that understands you — your skin tone, your occasion, your cultural context — before making a single recommendation.

## What we built

StyleMirror is an AI-powered personal stylist. You pick an occasion (Wedding, Festival, Office, Casual), upload a selfie, and within seconds you get:

• A skin analysis — undertone, skin type, radiance & moisture scores
• A virtual makeup try-on — skin-smoothing preview rendered on your actual face
• Curated jewellery & beauty picks — matched to your undertone and the occasion's colour palette

Everything is woven into one seamless flow, not three disconnected tools.

## How we built it

The frontend is a vanilla HTML/CSS/JS single-page app — five screens driven by a goTo() navigation function, with a glassmorphism dark-gold design system. A Node.js/Express backend proxies all Perfect Corp API calls so the API key never touches the browser.

The backend pipeline runs in parallel:

1. Upload the selfie to Perfect Corp via the S2S File Upload API
2. Fire Skin Analysis and Makeup VTO concurrently
3. Poll both tasks until task_status === "success"
4. Unzip the skin analysis result, parse score_info.json, and return structured JSON to the frontend Product recommendations are occasion-aware: each of the four occasions carries its own colour palette and curated accessory set, cross-referenced with the skin undertone returned by the API.

## Challenges

Async polling across two APIs in parallel was the trickiest engineering problem. Both tasks are fire-and-poll, and we needed both results before rendering — so we used Promise.all with individual exponential-backoff pollers rather than a single polling loop.

Undocumented VTO effect catalog — effects like lip_color and blush require shape.name and pattern values that aren't in the public docs. We reverse-engineered working configurations through trial and error, and ultimately shipped with skin_smooth as the confirmed-working effect while logging the others for a post-hackathon follow-up.

Designing for cultural diversity — generic Western beauty references don't map well to South Asian occasions. Building occasion-specific palettes (Vermillion + Marigold for Holi festivals, 22k gold tones for bridal) required research beyond the API docs.

## What we learned

Perfect Corp's S2S API is production-grade and fast — skin analysis + VTO in under 8 seconds end-to-end. The biggest unlock was treating the two APIs as a pipeline rather than sequential steps, which cut perceived wait time significantly.

Built With

• analysis
• css
• express.js
• html
• javascript
• makeup
• node.js
• s2s
• skin
• vto