Inspiration
Most skincare shopping starts the wrong way round: a shelf full of products, no idea which ones are actually doing anything, and a new purchase every time a new concern appears. We wanted the opposite — start from your skin, audit what you already own, and only buy the missing piece.
What it does
SkinCart turns a selfie into a shopping plan you can trust:
- Understand — the YouCam Skin Analysis API reads your photo and returns an overall score, skin age, skin type, and a score for every concern (in our sample run: 78/100 overall, skin age 27, Combination type, with Hydration, Texture and Redness as the top focuses), plus a skin map with 17 overlay layers.
- Audit — tell SkinCart what you already own, and it checks every product against your analysis: what it covers, what it duplicates, and what conflicts. An exfoliating toner and a retinol serum, for example, are never scheduled on the same night — they alternate as Night A and Night B.
- Meet Sage, your agent — Sage is a conversational buddy who explains every decision in plain words and shows her working. She greets you with your real top concern and score, walks through six decisions (read the analysis, audit the shelf, check conflicts, fill the gaps, check the budget, build the routine), and answers back when you push her: "Why this one?" gives the product, the gap, your concern scores and the recorded ingredient groups; "Make it cheaper" states the current total and works to a monthly budget in ₹, swapping to a cheaper option only when it passes the same checks — and saying so plainly when no cheaper option does. "Didn't suit me" removes a product and re-plans on the spot.
- Prove it's working — save scans over time and compare them in "Your skin story" (our sample profiles show Week 0 → Week 4: overall +5, Hydration 58 → 76). SkinCart never claims a product works without evidence: where the catalogue has no recorded evidence, it says so instead of inventing a claim.
How I built it
React + Vite + Tailwind on the front end, with an Express proxy that holds the YouCam API key server-side (never in the browser). The app runs in demo mode by default so the full flow can be explored with zero API units, and switches to live analysis when you turn demo mode off. Sage's agent is a transparent, deterministic rules engine — every recommendation traces back to a score, a shelf item, or a recorded ingredient group, which is why she can explain herself instead of asking you to trust a black box.
Challenges I ran into
Keeping every claim honest (prices marked illustrative where the catalogue has no live price, missing evidence stated rather than papered over), making a routine that respects real skincare conflicts, and making the agent feel like a companion without letting it invent anything — the chat UI sits on top of the same rules engine, not on a language model.
Accomplishments I am proud of
A full end-to-end flow verified by repeated QA: sample analysis, shelf audit, conflict-safe routines, budget swaps with correct maths, a feedback loop that visibly re-plans, a transcript that survives a reload, and a six-decision trace any judge can open and check.
What's next
Makeup virtual try-on for the recommended tinted product (the natural second half of this hackathon's two topics), real INR prices behind the price-check button, and weather/UV-aware routine tweaks.
Built With
- express.js
- javascript
- node.js
- react
- tailwind-css
- vite
- youcam-skin-analysis-api