01
Transparent by default
Every score, every time. You always see the full breakdown across all seven categories, not just a headline, because a number you can’t interrogate isn’t much better than a guess.
About SkinSense
Built to make skin advice legible.
SkinSense turns one photo into something you can actually read: a screening result, a confidence figure you can weigh, and a routine short enough to keep.
The problem
A search, a wall of contradictory advice, a shelf of products bought on a guess.
That’s how most people meet a skin problem. One source swears by an ingredient another warns against. A product is recommended without a word about what’s in it. It’s hard to know what you’re looking at, harder to know what to try, and hardest of all to tell whether it’s working.
Meanwhile, getting a dermatologist’s opinion can take time, and many everyday concerns never get one. People are left guessing, and guessing is expensive.
Our mission
Replace the first hour of guessing with something concrete.
A transparent screening result. An ingredient-led routine from products you can find at a pharmacy. A simple way to track your own progress. And a clear hand-off to a dermatologist the moment your results suggest you need one.
Principles
Four rules we build by.
- 02
Ingredient-led
Routines name the actives, not just brands: niacinamide, salicylic acid, azelaic acid, ceramides. Any product with the same key ingredients does the same job.
- 03
Escalates when it should
Severe results are flagged for a doctor, as are moderate results for rosacea, eczema and hyperpigmentation, and repeated severe check-ins.
- 04
Privacy first
Passwordless sign-in, optional profile questions, and only what’s needed stored in a private dataset. Photos are not sold or shared.
Under the hood
How the model works.
SkinSense uses a fine-tuned EfficientNet-B0, a compact convolutional neural network designed to be accurate for its size. It starts from a version pre-trained on general images and is fine-tuned on labelled skin photos across the seven categories.
- 1
Your photo
One face photo, uploaded or taken with your camera.
- 2
Prepared
Resized to 224 × 224 pixels and colour-normalised, the format the model learned from.
- 3
Classified
EfficientNet-B0 outputs a probability for each of the seven categories.
- 4
Explained
The top category is your result, its probability is your confidence, and a severity band is derived from it.
- 5
Actioned
The result maps to AM and PM routines, habit tips and, when warranted, a recommendation to see a dermatologist.
- Architecture
- EfficientNet-B0, fine-tuned
- Input
- One RGB photo at 224 × 224
- Output
- Probabilities for 7 categories
- Result
- Top category + its confidence
- Severity band
- Derived from that confidence
- Doctor flag
- Severe, or moderate for rosacea, eczema, hyperpigmentation
Limits
What it can’t do, and why that matters.
Being clear about the limits is part of being useful. Please read these before relying on a result.
Not a diagnosis. SkinSense is an educational screening tool and is not a medical device. It does not diagnose, treat or cure any condition. Always consult a qualified dermatologist or physician about a skin concern, and seek care promptly if your skin is painful, bleeding, spreading or changing rapidly.
It only knows seven categories
Anything else, such as a mole, an infection, psoriasis or a rash, is still sorted into whichever of the seven it most resembles.
It reads a photo, not your skin
Lighting, camera quality, makeup, filters and angle all change the result. Poor photos give unreliable readings.
Performance can vary across skin tones
Redness and pigmentation look different on different skin, and a model is only as broad as the photos it learned from.
Severity isn’t a clinical grade
The band reflects how strongly the photo matches the top category, not how serious your condition is.
It can’t ask questions
A dermatologist considers your history, symptoms, how your skin feels and how fast it’s changing. A photo can’t capture any of that.
It isn’t a medical device
SkinSense does not diagnose, treat or cure any condition. It’s a starting point for understanding, not a substitute for care.
What’s next
Where we’d like to take it.
Intention
Broader, fairer evaluation
We want to better understand how the model behaves across skin tones and lighting conditions, and improve where it falls short.
Intention
Better photo guidance
We’d like the capture step to spot poor lighting or focus before a photo is analysed, so fewer readings end up low-confidence.
Intention
Self-serve data controls
We aim to let you download or delete your own data from inside the app, without needing to email us.
Intention
Easier hand-offs
We’d like to make it simple to bring your check-in history to a dermatologist appointment.
One photo. A clear reading. A routine you can keep.
Free to use. Sign in with your email, verify a 6-digit code, and analyse your first photo.
Educational screening, not a diagnosis.