Proof

Is your skincare actually working? Weekly YouCam skin scans show which products help, which irritate, and what to stop paying for, with a face map and your goal skin.

Comment

Try it: https://proof.revenuerush.in · judges' login [email protected], code 261101 (no email is sent for this account)

How to test (3 minutes)

  1. Open https://proof.revenuerush.in, click Sign in, enter [email protected], then the code 261101.
  2. The account opens with 7 weeks of scans and 4 products. Read the report: the summary, what you could stop spending, the routine check.
  3. Your face map: tap any concern dial to see where YouCam detected it on the face.
  4. Your skin, goal: drag the slider to compare today with the AI Skin Simulation.
  5. Product verdicts and Shop the fix: each verdict lists its evidence; add a suggested swap with one click.
  6. Weekly scan: click Use camera (or upload a front-facing selfie, face filling the oval) and Scan my skin. Watch the steps stream in; the face map, goal and verdicts update in about 20 seconds. No camera? Add a simulated scan runs the same pipeline without a photo.
  7. Add a product → Snap the label: upload a photo of any skincare bottle and Amazon Nova fills in the form.
  8. Share my progress draws the shareable card. Demo profile resets the account to the starting data at any time.

Inspiration

People wonder about their skin at very specific moments: right after a breakout, or standing at the mirror deciding whether a product is working. Most of us spend hundreds a month on serums and creams and have no idea which one is helping and which one is quietly irritating our skin. We quit what works too early and keep paying for what doesn't. Proof turns that guess into evidence.

What it does

Proof is a report card for your skincare.

  • Weekly scan. A ten-second selfie from the live camera (cropped to a face guide) or an upload. The YouCam Skin Analysis API scores nine concerns: pores, redness, acne, moisture, texture, wrinkles, spots, radiance and oil control, plus skin age.
  • Face map. YouCam's per-concern detection masks show where each concern sits on your face: shine across the T-zone, fine lines under the eyes, a spot on the cheek. Tap any concern.
  • Your skin, goal. YouCam AI Skin Simulation shows your own face with your most pressing concern improved, as a drag-to-compare before/after: the result to aim for.
  • Verdicts with evidence. Proof compares your skin in the weeks before and after each product started, works out which product caused which change, and gives every product a verdict: Keep, Pause (irritation), No effect, or Too early, with the numbers behind it and how much you'd save each month.
  • Routine check from day one. Flags clashing actives (retinol with vitamin C, acids with retinol) and weak concerns nothing in your routine targets, so the first scan is already useful.
  • Shop the fix. A gentler swap for each irritating product (bakuchiol instead of retinol), a different active for products that did nothing, and a pick for untreated concerns, each added to your routine in one click.
  • Snap the label. Photograph a bottle and Amazon Nova reads the name, key actives and price.
  • Share card. A 1080×1350 progress card with your score, change since week one, verdicts, savings, and today next to your goal.

How we built it

  • YouCam API v2 from a worker Lambda: upload → skin-analysis task (9 SD concerns, JSON output, mask overlays) → poll; then skin-simulation on the concern that matters most. Mask and simulation links expire in two hours, so the worker copies them to S3.
  • Attribution engine: for each product, scans from the 21 days before it started vs. days 4–28 after; a decline is blamed only on the product whose window shows it most strongly, so one irritating product doesn't make everything look bad. Products used before tracking began are judged on the trend of their target concerns.
  • Agent on Amazon Nova 2 Lite (Bedrock Converse, tool use with schema validation and retries). Proof's analysis decides every verdict and every number; the agent explains them in plain language. A rules fallback keeps reports working if the model is unavailable.
  • Serverless AWS, one CDK stack: CloudFront + S3 site, Cognito passwordless sign-in (custom auth, 6-digit email codes sent through SES), AppSync GraphQL, an API Lambda, SQS → worker Lambda for scans, reports and label reading, DynamoDB, and S3 for photos. Progress streams to the browser over an AppSync subscription filtered to the owner. Bedrock and SES run through invoke-only roles in a second account.
  • Front end: React + Vite, an editorial beauty look (Bodoni display, dark plum and iridescent palette), camera capture with a face guide, a canvas-drawn share card.
  • Demo video generated from the live app with Playwright, narrated with Amazon Polly.

Challenges we ran into

  • Faces too small. Laptop webcams frame a lot of room; YouCam rejects small faces (error_src_face_too_small). The camera now crops to the face oval before upload.
  • Pre-signed S3 links. YouCam's servers couldn't fetch them (error_download_image), so the worker uploads the image bytes through YouCam's file API instead.
  • Keeping an LLM honest. Early on the model sometimes said a product was helping when the numbers said otherwise. We moved every fact (verdicts, evidence, routine notes) into deterministic code and let the agent only explain.
  • Attribution. Several products start at different times; a naive before/after blames every product for one product's irritation. Windowing and "worst-window ownership" fixed it.

Accomplishments that we're proud of

  • Three YouCam capabilities working together (scores, detection masks, skin simulation) in one flow that a first-time user understands immediately.
  • A full production deployment with real sign-in, private per-person data, live progress, and a daily scan limit that protects API units.
  • Verdicts a user can trust, because every number on screen comes from their own scans.

What we learned

The value isn't in a single skin score; it's in the change over time and in tying that change to a decision. Showing where a concern sits and what the goal looks like makes the numbers feel real.

What's next for Proof

  • Weekly scan reminders, and progress toward the simulated goal
  • Real retailer catalogs and affiliate links behind Shop the fix
  • Patch tests: try a product on one side of the face and let Proof compare the two halves
  • Sharing a report with a dermatologist

Built With

Share this project:

Updates

Submission history