GlowCoach

One selfie becomes a skin plan you can shop: YouCam AI skin analysis, a routine for your budget, a matched foundation shade, makeup try-on, and a preview of your skin goals.

Comment

Inspiration

Shopping for skincare online is guesswork. You read ingredient lists, guess your skin type, and buy a foundation shade from a swatch on someone else's screen. I wanted one selfie to answer the questions a good skincare advisor would ask, and then let you shop the answer.

What it does

GlowCoach turns one selfie into a skin plan you can shop:

  • Scan: YouCam AI Skin Analysis scores 8 concerns (breakouts, pores, texture, redness, oil, hydration, fine lines, dark circles). Tap a concern to see where it was found.
  • Plan: a morning and night routine built for your lowest scores, the goals you pick, and your budget. Every product says why it was chosen.
  • Start slow: a 4-week plan adds the basics first, then treatments, then strong actives like retinal or BHA two to three nights a week. If the scan shows redness, actives start more slowly. Starting everything at once is the usual reason people get irritated, quit a routine, and return the products.
  • Try on: Skin Tone Analysis finds your skin color, GlowCoach matches the nearest foundation shade and undertone, and Makeup Virtual Try-On applies three looks to your photo with a before/after slider.
  • Preview: Skin Simulation shows your photo with the routine's target concerns improved, clearly labeled as an illustration.
  • Shop and track: add the routine or a look to your bag, and rescan over time to see your scores change.

How I built it

A static web app (HTML, CSS, JavaScript, no framework) with Node.js serverless functions. The browser never sees the API key: it sends the photo to my server, which uploads it to YouCam, starts tasks from an allow-list, and polls for results. Skin analysis and tone analysis run in parallel; try-on and simulation only run when asked, to save units. The coaching logic (priorities, routine picking, the 4-week start plan, CIE Lab color matching) is plain, tested functions: 34 automated tests plus browser tests.

Challenges

  • Units are precious, so I added a per-visitor and daily spending guard, and a sample mode recorded once from real API results so anyone can explore the full journey without spending units.
  • Result links expire after 2 hours, so the sample results are saved with the app.
  • Undertone from a single hue cut-off misread very light and deep skin, so I take the undertone from the nearest matched shade instead.

What I learned

How to design around asynchronous AI tasks (upload, start, poll) and how much a clear "why this product" builds trust in a recommendation.

What's next

Plug in a real retailer's catalog, add more looks, and send scan reminders so progress tracking becomes a habit.

Built With

Share this project:

Updates

Submission history