DermisIQ

DermisIQ is a clinical-grade mobile AI skin concierge that bridges the gap between computer-vision dermatological diagnostics (Perfect Corp YouCam AI API) and safe chemical skincare formulation.

Comment

DermisIQ — Melanin-Smart Skin AI & Molecular Ingredient Clash Copilot

A clinical-grade mobile application pairing Perfect Corp YouCam AI diagnostics with an active ingredient contraindication engine to prevent chemical barrier breakdown and reduce shade-mismatch eCommerce returns from 35% down to <2%.

💡 Inspiration

Skincare today suffers from two massive, interconnected problems:

  1. The Active Skincare Epidemic & Chemical Burns: Consumers are layering clinical-strength actives—retinoids, AHA/BHA exfoliants, pure L-ascorbic acid, and benzoyl peroxide—without knowing their skin's barrier status. The result is chronic stratum corneum damage, microvascular erythema, and irritant contact dermatitis.
  2. Melanin Inequity & The Rebound PIH Paradox: Skincare advice has historically been benchmarked against fair phototypes (Fitzpatrick I–II). However, in melanin-rich skin (Fitzpatrick III–VI: South Asian, Black, Hispanic, Middle Eastern), aggressive rapid-penetration acids (like 70% unbuffered glycolic acid) inflame melanocytes, triggering stubborn Post-Inflammatory Hyperpigmentation (PIH) that takes months or years to resolve.
  3. eCommerce Shade Mismatch & Return Waste: Tinted mineral SPFs and complexion products suffer from a 30% to 40% Return-to-Origin (RTO) rate due to inaccurate shade selection.

I built DermisIQ to replace guesswork with real-time biometric telemetry: automating phototype identification, predicting chemical clashes before products touch skin, and ensuring shade-matched photoprotection.

🔬 What We Learned & Mathematical Foundations

During development, we explored the biophysics of epidermal optics and molecular transdermal penetration kinetics:

1. Fickian Transdermal Diffusion Rate

Active ingredients diffuse through the stratum corneum lipid matrix following Fick’s First Law:

$$J = -D \cdot \frac{dC}{dx}$$

Where:

  • (J) is the transdermal flux rate ((\mu\text{g}/(\text{cm}^2 \cdot \text{hr})))
  • (D) is the solute diffusion coefficient, which scales inversely with molecular weight ((\text{MW}))
  • (\frac{dC}{dx}) is the concentration gradient across stratum corneum thickness (x)

Small molecules like Glycolic Acid ((76.05\text{ g/mol})) penetrate rapidly, triggering acute neurosensory stinging in reactive barrier states. In contrast, large-molecule polyhydroxy acids like Gluconolactone ((178.14\text{ g/mol})) and Azelaic Acid ((188.22\text{ g/mol})) diffuse gradually, preventing melanocyte panic.

2. Stratum Corneum Barrier Health Index (BHI)

DermisIQ synthesizes a unified Barrier Health Index (0–100) from multispectral YouCam metrics:

$$\text{BHI} = w_1 \cdot M_{\text{hydration}} + w_2 \cdot (100 - E_{\text{erythema}}) + w_3 \cdot (100 - P_{\text{porosity}}) - \lambda \cdot C_{\text{clash}}$$

Where:

  • (M_{\text{hydration}}) is optical stratum corneum water retention
  • (E_{\text{erythema}}) is microvascular erythema redness intensity
  • (P_{\text{porosity}}) is follicular pore dilation
  • (\lambda \cdot C_{\text{clash}}) is a penalty term deducted for active ingredient clashes

🛠️ How We Built DermisIQ

DermisIQ was engineered as a high-performance cross-platform application using React Native, TypeScript, and Perfect Corp YouCam S2S AI API Gateway:

1. Five Live YouCam S2S Cloud Neural Endpoints

  • POST /s2s/v2.0/file: Secure S3 pre-signed binary upload slot for multi-megabyte portrait captures.
  • POST /s2s/v2.0/task/fitzpatrick-scale-analyzer: Real-time Roman scale detection (Type I to Type VI), eliminating unreliable self-reporting quizzes.
  • POST /s2s/v2.0/task/skin-tone-analysis: Hyperspectral undertone mapping (e.g. Warm Olive, Golden, Neutral Deep) and localized hex swatches (#A26E4F).
  • POST /s2s/v2.0/task/face-attr-analysis: Landmark alignment, 26 facial proportion ratios, and biological skin age estimation.
  • POST /s2s/v2.0/task/skin-analysis: 8-parameter dermal health mapping across erythema, hydration, PIH spots, texture, pores, and oiliness.

2. 6-Axis Interactive Facial Radar Visualizer

A custom SVG biometric radar maps 6 facial diagnostic zones (Forehead, Left Cheek, Right Cheek, T-Zone, Eye Contour, Chin) with dynamic coordinate targeting and calibrated point values.

3. Molecular Clash & Contraindication Matrix

An offline-capable rule engine tracks pH incompatibilities, phototoxicity (e.g., Retinoids + direct UV without SPF), and melanin sensitivity. It flags warnings such as:

  • Direct Acid Stacking: AHA + BHA simultaneous application.
  • Oxidation Clashes: Benzoyl Peroxide + Pure L-Ascorbic Acid neutralization.
  • Melanin Warning: High-percentage glycolic peels flagged for Fitzpatrick IV–VI with gentle alternatives (Azelaic Acid 10%, PHA) automatically recommended.

4. Resilient Offline Baseline & Token Conservation

Recognizing limited API token quotas during production evaluations, we engineered:

  • In-Memory Biometric Scan Cache: Re-scanning or viewing previously verified photos consumes 0 API tokens.
  • Sequential Fail-Fast Pre-Validation: Evaluates photo resolution ((\ge 480\times 480\text{px})) and face framing on-device before firing network calls. If an image is invalid, execution halts before calling secondary endpoints—saving 75% of token burn.
  • Failover Telemetry: Airplane-mode resilience using calibrated dermatological baselines if offline or credentials exhaust.

🧗 Challenges We Faced

  1. Strict Neural Resolution Constraints: YouCam rejects images outside the (480\text{px} \le \text{dim} \le 4096\text{px}) window or when the face occupies (< 60\%) of the frame (error_below_min_image_size, error_src_face_too_small). We built client-side image resizing and pre-upload aspect validation with real-time toast notifications guiding the user.
  2. Eliminating UI Overlay Blocking: We decoupled the live camera modal from tab screens and resolved navigation backdrop conflicts by mounting a unified modal at the root level.
  3. Persisting Cold-Boot Biometrics: Ensuring camera captures, inferred phototypes, and 6-axis data points survive app restarts by implementing AsyncStorage state hydration with zero flicker.

🔮 What's Next for DermisIQ

  • Dermatologist Tele-Triage Export: One-tap clinical PDF reports mapping 30-day barrier progression and clash logs.
  • Barcode & Ingredient List OCR: Live camera scanning of cosmetic ingredient labels (INCI) to detect clashes in store aisles.
  • Smart Dispensers Integration: IoT pairing with smart vanity dispensers to calibrate serum doses based on daily YouCam barrier scores.

Built With

Share this project:

Updates

Submission history