ProofLoop

A fair test for your skincare. Measure your skin with the YouCam AI Skin Analysis API, choose one product for one goal, and check back weeks later. ProofLoop tells you whether a score moved further than normal scan-to-scan variation — or says plainly that it can't tell yet.

Built for the YouCam API Skin AI & eCommerce VTO Hackathon (Skin AI track).

Why

People wonder whether a product is working while standing in front of a mirror — a bad instrument with different light every morning. Skin scanners give a number; shops give a product; nobody tells you whether the next number is actually different from the last one. ProofLoop closes that loop honestly:

1. Measure — a framed selfie goes to YouCam Skin Analysis (8 concerns, scores 1–100, higher = better-looking).
2. Choose one focus — each focus is tied to one measured score.
3. Pick a product — deterministic matching from a catalogue, with every reason shown and no fake percentages. Optional goal preview from YouCam AI Skin Simulation.
4. Fair-test plan — when the trial starts, ProofLoop fixes what will count: the scores that would mean up or down, the band (measured, or assumed and set to the strictest measured), and a day-28 check-in. Focuses at the top of the scale, where a gain can't register, are flagged before you choose them.
5. Instrument check (optional) — before the first check-in, take a second photo of unchanged skin. ProofLoop compares the two scans; if they disagree by more than the standard band, your band widens to fit them (max(band, difference + 2)). It never narrows, and it's sealed into the plan at the first check-in.
6. Check in — a free capture-match check compares the new photo's lighting and camera with the baseline before any units are spent (and refuses the baseline photo itself, which would score identically); the scan is then judged against the plan. Outcomes: measured change, no clear change, or inconclusive — and "early look" before the planned date. Each verdict ends in one next step that follows from the plan: keep going, finish the test, consider switching (back to same-goal matches), pause and review, or retake.

How YouCam is used

API	Endpoint	Where
File upload	POST /s2s/v2.0/file/skin-analysis + presigned PUT	every live scan / preview
AI Skin Analysis	POST /s2s/v2.0/task/skin-analysis, GET …/{task_id} (format: json, 8 SD concerns, 12 units)	baseline, instrument-check and check-in scans; per-concern detection masks shown over the user's photo
AI Skin Simulation	POST /s2s/v2.0/task/skin-simulation, GET …/{task_id} (4 units)	optional goal preview on the product page

All of the above were verified against the live API through ProofLoop's own client — see docs/youcam-integration.md for evidence, costs and the repeatability study.

Run it locally

Requirements: Python 3.12, Node 22.

# backend (http://localhost:8000)
cd backend
python3.12 -m venv venv
./venv/bin/pip install -r requirements-dev.txt
cp .env.example .env          # add YOUCAM_API_KEY for live scans; leave blank for sample data only
./venv/bin/uvicorn app.main:app --reload --port 8000

# frontend (http://localhost:3000)
cd frontend
npm install
npm run dev

Without a key, everything except live scans and previews works, and live scans fail with a clear message instead of inventing results. Try with sample data replays a recorded YouCam response, labelled as such.

Tests

cd backend && ./venv/bin/python -m pytest          # 91 tests, no network; 2 opt-in live tests skipped
cd backend && TEST_DATABASE_URL=postgresql://… ./venv/bin/python -m pytest   # same suite on PostgreSQL, schema via Alembic
cd backend && ./venv/bin/ruff check . && ./venv/bin/ruff format --check .
cd frontend && npm run lint && npm run typecheck && npm run build
cd frontend && npm run test:e2e                     # Playwright, desktop + mobile, axe; isolated keyless backend

Live checks (spend units only when asked): YOUCAM_LIVE=1 ./venv/bin/python -m pytest -m live.

Deploy

Hosted on free tiers: Next.js on Vercel (Hobby), FastAPI on Render (Free web service), PostgreSQL on Neon (Free). Schema changes go through Alembic (backend/migrations). Runbook, settings and limits: docs/deployment.md.

Privacy

Photos are cleaned (metadata stripped) and sent to YouCam, never stored by ProofLoop. The server keeps scores, capture metadata and trials under an anonymous session token; "Delete all my data" removes them immediately. ProofLoop is not a medical device and never attributes a change to a product.

Docs

• Innovation pass (2026-10-19): audit, research, decision, demo plan
• Product strategy · Competitive analysis
• Architecture · Decisions · Audit of the previous MVP
• YouCam integration & evidence
• Design direction
• Deployment
• Demo script · Submission copy · Final audit

The product catalogue is fictional sample data created to demonstrate matching.