Our Methodology

When you upload or capture a photo, it is transmitted over HTTPS to our vision pipeline. We do not store the raw image unless you explicitly choose to save the report to your account. The image is normalized for exposure, white balance, and orientation, and we run a quick quality check: is the face in frame, is there enough light, is the image sharp enough for the model to read texture. If any check fails, you get a prompt to retake rather than a garbage result.

The image is passed to a multimodal vision model that has been prompted with a structured rubric drawn from dermatology and trichology literature. The model returns discrete observations across roughly 40 visible signals: skin tone evenness, visible pore density, sebum distribution, erythema, texture, fine-line depth, undereye pigmentation, lip hydration cues, visible hair density at hairline and crown, miniaturization patterns, scalp visibility, and overall grooming state. It does not attempt to diagnose medical conditions. It also does not attempt to estimate weight, race, sexuality, or emotional state, and we actively filter those categories out of its output.

The discrete observations are turned into four numeric scores (Skin, Hair, Diet cues, Wellness cues) on a 0-100 scale. Each score is a weighted composite; the weights were set by our reviewing dermatologist and dietitian, not by the model itself. Diet and Wellness scores are inferred cues (undereye congestion, skin dullness patterns, hair fragility) rather than direct measurements, and they are labeled that way in the UI so you know what the number actually means.

The scores and observations feed into a plan generator that composes a personalized routine from a curated library of interventions (product categories, ingredient recommendations, dietary swaps, sleep and stress tactics). Every recommendation in that library has a source note tied to a study, a guideline, or an expert-review decision. Products are recommended by category first (for example, "a fragrance-free ceramide moisturizer"), with named examples surfaced afterward only when our reviewer has independently approved them.

When you scan again in the future, the new scores are compared against your baseline under matched lighting conditions where possible. We show trends rather than absolute numbers because absolute numbers are noisy day to day; what matters is the direction of travel over 4 to 12 weeks.

Photos are transmitted over TLS, processed in memory, and discarded unless you save the report. We do not sell data. We do not train our vision model on your face. Saved reports live in your account and can be deleted at any time from the Profile page. Full details are in our privacy policy.