SkinProof

Proof for every skin treatment: clinic peel, prescribed cream or serum. Guided WhatsApp selfies + YouCam Skin AI turn real results into evidence clinics, brands and shoppers trust.

Comment

Inspiration

Every skin decision in India starts with the same question: will this actually work for my skin? A patient paying for six acne-peel sessions can't see whether she's improving until it's too late, so many stop halfway. A clinic knows its protocols work but can only show a few hand-picked before/after photos. A skincare brand claims "fewer dark spots in 4 weeks" based on a small panel few shoppers believe.

All three are missing the same thing: consistent, measured outcomes. We already run ClinicFlow, a WhatsApp front desk for Indian skin and laser clinics, so we had the channel patients actually use. YouCam Skin AI gave us the measurement.

What it does

SkinProof treats every skin intervention the same way, whether it's a clinic procedure (peel, laser, facial), a prescribed cream, an over-the-counter product, or a combination (a "regimen"):

  1. Guided baseline scan on WhatsApp or web. The patient sends a selfie. A photo coach turns YouCam error codes (face too small, too dark, out of frame) into simple retake instructions in English and Hindi, so scans are comparable week to week.
  2. Skin AI + Fitzpatrick typing. YouCam Skin Analysis scores the concerns relevant to the treatment (acne, age spots, redness, pores, texture and more) and returns mask overlays. The Fitzpatrick Skin Type API records skin type.
  3. Doctor stays in control. The doctor's pre-read shows overlays and Fitzpatrick type. For darker skin types, where laser settings need extra care, laser slots open only after the doctor reviews the scan. SkinProof flags; the doctor decides.
  4. Personal progress. Each re-scan before a session shows the patient her change from baseline, as a report card and a Hindi voice note.
  5. Outcomes across patients. Each scan adds to results for that protocol: share of patients improved, average change, dropout. Clinics publish proof pages for their protocols. Brands use the same engine to run real-world product trials and get a verified-claim page shoppers can scan from the pack.

Guardrails: results are reported per regimen, never "the cream did it." Prescription-cream results stay private to the doctor; only clinic protocols and over-the-counter products get public pages. SkinProof is cosmetic skin measurement, not diagnosis.

How we built it

  • YouCam APIs: AI Skin Analysis (file → task → poll; raw_score stored for tracking, ui_score shown to patients, mask overlays for report cards) and AI Fitzpatrick Skin Type Analysis.
  • Channel: WhatsApp via AWS End User Messaging Social → SNS → Lambda, plus a mobile web scan page with a face-outline camera guide.
  • Agent: an agent with a new skin_scan tool alongside ClinicFlow's booking tools; explains results in English or Hindi and hands off to booking.
  • Data: DynamoDB single-table model (Intervention, Regimen, Cohort, Scan); selfies and masks in S3 with consent and delete-on-request.
  • Voice: Amazon Polly for Hindi result voice notes; Amazon Transcribe for voice input.
  • Dashboards: clinic outcomes per protocol, brand trial view, public proof pages.

What we significantly updated during the submission period

SkinProof is built on our existing ClinicFlow WhatsApp front desk. New during this hackathon: YouCam Skin Analysis + Fitzpatrick integration, photo coach, mask-overlay report cards and Hindi voice results, Fitzpatrick-gated laser booking, the intervention/regimen/cohort model and outcome statistics, clinic/brand dashboards, proof pages, and the web scan page. [TBD: confirm after build]

Challenges we ran into

  • Making selfies comparable: lighting and framing change scores more than a week of treatment does, so we built a photo coach and track raw_score.
  • Honest attribution: patients rarely use one thing at a time, so we report per regimen.
  • API unit budget: developed on SD with few concerns and cached responses; saved HD for the demo.

Accomplishments that we're proud of

  • One measurement engine for clinics, brands and patients.
  • A doctor-in-the-loop safety gate for laser treatment on darker skin types.
  • Results as Hindi voice notes for patients who'd rather listen than read.

What we learned

  • The useful output of Skin AI isn't a single score, it's the change in that score under controlled conditions.
  • In India the channel matters as much as the model: WhatsApp and voice reach patients a web app never would.

What's next for SkinProof

  • A 4-week pilot at a partner clinic in Meerut to replace simulated cohort data with real outcomes.
  • Recruiting brand trial participants from clinic patients.
  • Comparing regimens across clinics (peel vs cream vs both) for similar skin types.

Note on demo data: historical cohort results in the dashboard are simulated seed data, labelled in the UI. Live scans in the video are real YouCam API calls.

Built With

Share this project:

Updates

Submission history