Start AuditThesciencebehindeveryQutisaudit.
Every ingredient score, every audit rule, every flag is peer-reviewed, dermatologist-validated, and calibrated for Fitzpatrick IV–VI skin. This is how we know our recommendations are safe to ship.
Five stages, in plain language.
No black box. From the moment your photo lands in our inbox to the moment your audit report is signed off — here's what happens.
- Stage 01
WhatsApp input
Image, product label, or text concern arrives via the WhatsApp Business API. End-to-end encrypted in transit, ephemeral by default. The user never leaves their messaging app.
- Stage 02
Face & skin region detection
A Multi-task CNN (MTCNN) detects facial regions and isolates the skin areas relevant to the concern. Non-skin pixels are masked out before any classification step runs — reducing background noise and false positives.
- Stage 03
AI vision analysis
A medically-tuned vision-language model produces a structured assessment: condition flags, severity scoring, and tone-aware observations calibrated for the patient's Fitzpatrick type.
- Stage 04
Rules engine + ingredient database
Each detected condition is cross-referenced against our dermatologist-curated rules database. 80+ safety rules for melanin-rich skin determine recommended actions, ingredient swaps, and escalation criteria.
- Stage 05
Dermatologist review & audit report
Every audit is sign-off-reviewed by our licensed dermatologist within one business day. Edge cases are escalated to in-person consultation; the final report is delivered back to the user inside the same WhatsApp thread.
What makes the audit calibrated.
Fitzpatrick-specific training data.
We curated 12,000+ annotated images of skin conditions on Fitzpatrick IV–VI from East African clinical partners and de-identified consented user data. Standard public datasets remain heavily skewed toward lighter skin; ours doesn't.
Dermatologist-reviewed rules engine.
Every recommendation passes through 80+ safety rules co-authored with our dermatology team. PIH risk, irritation thresholds, sunscreen interactions, and pregnancy-safe substitutes are all hard-coded — not learned probabilities.
Ingredient database for East African retailers.
We track 5,000+ products available across Kenya, Tanzania, and Uganda with batch-level ingredient lists and price tiers. When we recommend a swap, we recommend something you can actually buy this week.
What we're reading, what we're publishing.
Accuracy gaps in AI dermatology for darker skin tones.
A 2024 benchmark comparing top commercial AI dermatology tools across Fitzpatrick I–VI. Top performers showed 23–34% accuracy drops on Fitzpatrick V–VI relative to lighter skin tones.
Read studyClinicalPIH prevalence and recurrence in Fitzpatrick IV–VI.
A two-year follow-up of 480 patients across three East African clinics, tracking post-inflammatory hyperpigmentation recurrence patterns after routine-led interventions.
Read studyMethodologyIngredient safety considerations for melanin-rich skin.
A systematic review of 27 cosmetic actives commonly recommended in mass-market routines, scored for PIH risk and irritation potential on dark skin.
Read studyCare deliveryWhatsApp as a healthcare delivery channel in East Africa.
Mixed-methods research across four East African markets on patient preferences for chat-first health interactions versus traditional clinic and telehealth modalities.
Read study“We can't keep building dermatology AI that works brilliantly on the skin tones the original training data over-represents — and then act surprised when it fails the rest of us. Building Qutis was about closing that gap. Case by case. Ingredient by ingredient.”
