Inspiration

43% of beauty shoppers cite greenwashing as a purchase barrier. Every serum claims “reduces wrinkles 15% in 8 weeks,” but skincare efficacy takes 4–12 weeks to manifest — there is no “try before you buy,” only hope-based purchasing and some of the highest return rates in e-commerce. Brands can't prove claims on a customer's specific skin, and consumers can't verify anything beyond marketing copy. We built Verified by YouCam to turn every consumer into their own n=1 clinical trial.

What it does

Verified by YouCam is a beauty efficacy verification platform that transforms "clinically proven" into "proven on your skin":

  1. Scan: A guided capture ritual (on-device lighting, sharpness, face QA, granular consent, EXIF stripped) measures 16 skin metrics via YouCam Skin Analysis, plus skin tone colors and facial attributes.

  2. Analyze: Metrics are structured into a longitudinal time-series. Deltas are computed against repeated baseline replicates with a per-metric session-noise model; changes below 2× noise are honestly flagged, never celebrated.

  3. Verify: Marketing claims are parsed and validated against a 20-ingredient efficacy knowledge graph (concentration-blended, synergy-capped, antagonism-aware) and k-anonymized cohort aggregates. Users get personalized p10/p50/p90 outcome bands before they buy, and an Efficacy Certificate at week 12 — signed by data, not the brand. Brands receive a read-only PDP badge (“73% of 2,847 verified users improved by week 8”) via a single <script> tag.

Our signature UI moment is the TrendBand: the cohort prediction band rendered as a soft ribbon with the user's actual progress line drawn over it — you vs. the prediction.

How we built it

We integrated 9 YouCam APIs into a comprehensive web platform:

  • Skin & Facial Analysis: skin-analysis v2.1 (all 16 HD actions with overlay masks) for baseline and 12-week longitudinal tracking, skin-tone-analysis for tone color palettes, and face-attr-analysis (high strictness) for facial ratios and cohort matching.

  • Hair Diagnostics: hair-density-detection, hair-type-detection, hair-frizziness-detection, and hair-length-detection for hair-care claim verification across 3-photo flows.

  • VTO Bridge & Credit Ops: makeup-vto / look-vto for post-certificate "projected look" previews, alongside client/credit + credit/feature-cost for strict credit reserve guardrails.

Architecture Stack:

  • Frontend/Backend: Next.js 15 app (apps/web) covering landing, scan, results, journey, dashboard, claims, certificate, verification, demo PDP, API docs, and credit ops.

  • Domain Engines: Core TypeScript packages (@packages/core-youcam, core-metrics, core-claims, core-predict, db, ui) powering a typed client mirroring the official Skill contract (upload → task → poll), a serverless-native fan-in poll architecture, an MCP server (youcam-verified, 7 agent tools), and a Drizzle-modeled data layer.

  • Dual-Engine Execution: Features a zero-config runnable Sandbox mode with a clearly labeled Sandbox/Live factory switch for deterministic testing, which routes to the live YouCam API when YOUCAM_API_KEY is provided.

Challenges we ran into

  • Eliminating False Positives & Noise: Real-world selfie analysis introduces session noise. We engineered a per-metric session-noise model to filter out natural variance and honestly flag changes below 2× noise instead of prematurely celebrating them.

  • Complex Ingredient Interactions: Building a 20-ingredient efficacy knowledge graph that accurately accounts for concentration blending, synergy caps, and ingredient antagonisms without giving misleading predictions.

  • Async API & Guardrails Handling: Ensuring proper execution of YouCam's async task model (dispatch → poll same task_id without re-dispatching) while managing strict credit guardrails and respecting ui_score semantics without rendering raw scores.

Accomplishments that we're proud of

  • Complete 9-API Integration: Successfully orchestrating 9 separate YouCam APIs into a unified, seamless user flow spanning skin, facial, hair, VTO, and credit endpoints.

  • "Clinical Luxury" Design System: Crafting an Aesop × Apple Health visual identity featuring our custom Radar16 morphing chart, animated VerdictSeal, TrendBand visualization, WCAG-AA dark-first theme, and spring-based motion grammar.

  • Honesty Engineering: Building a platform that isn't afraid to render an UNVERIFIABLE or UNSUPPORTED verdict on a product claim, giving users true scientific credibility.

What we learned

  • Honesty Builds Trust: In e-commerce retail, a platform that occasionally tells a user not to buy a product (when a claim is unsupported for their specific profile) makes its positive recommendations exponentially more valuable.

  • API Precision: Respecting strict API semantics (e.g., deriving undertones client-side as labeled estimates from raw tone colors, or distinguishing HD/SD pipelines) is critical for medical/clinical-grade UX integrity.

What's next for Verified by YouCam

  • Expanding the Knowledge Graph: Scaling the ingredient graph beyond 20 core ingredients to cover thousands of INCI formulations.
  • Expanded Brand Integrations: Rolling out self-serve SDKs and Shopify app blocks for seamless PDP badge embedding across e-commerce storefronts.
  • Live Brand Dashboard: Launching real-time cohort analytics, formulation insights, and claim-compliance reporting tools for beauty brand research teams.

Built With

  • nextjs
  • vercel
  • youcam
Share this project:

Updates

Submission history