Eczema Vision: Open-Source, In-Browser Eczema Detection from Skin Images

Nathan Francis · built under 1:1 engineering mentorship with Jeffrey Moon (Curious Cardinals) · code: github.com/nathanfrancisk-lang/dermatology_vision

Overview. Eczema affects over 70 million children worldwide, yet in the communities where FutureBright runs health camps, a dermatologist's opinion can be hours away. Eczema Vision (2026–present) is an open-source binary image-classification model trained to screen photographs of skin for signs of eczema: a first-pass triage aid, built from the same research program as my literature review and Drosophila work. The project covered the full applied-ML pipeline: image preprocessing, data labeling, model training, and performance evaluation.

Try it

Screen a photo

Pick or drop a photograph of the skin area. Processing happens locally on your device, nothing is uploaded anywhere.

Drop a skin photo here, or click to choose JPG or PNG · processed on-device · never uploaded

Analyzing…

Try another photo

Model weights are being published. The demo interface is ready, and the trained model file is on its way to this page. In the meantime, the code is on GitHub.

checking for model…

How it works

The model is a convolutional neural network trained for binary classification on labeled photographs of eczematous and healthy skin. Images are resized and normalized in the browser, then passed through the network with TensorFlow.js; the output is a single probability that the photograph shows eczema. Because inference runs client-side, the demo is private by construction (a property that matters for medical images) and works even on modest connections once the page has loaded.

The project was built end-to-end during a Curious Cardinals mentorship with Jeffrey Moon: assembling and labeling the dataset, preprocessing images, training and iterating on the classifier, and evaluating its performance. It connects directly to the field: FutureBright's health camps screen hundreds of patients in a day, and low-cost first-pass triage tools are one way that reach can grow.

Links

• Code, github.com/nathanfrancisk-lang/dermatology_vision (MIT)
• The research it belongs to, nathanfrancis.org § 3: Research
• The field program, FutureBright Inc.