AcneScan

Inspiration

Acne is one of the most common skin conditions worldwide, but access to dermatologists is slow, expensive, and intimidating—especially for students and teens. We wanted a way to quickly understand what kind of acne you might be dealing with, get plain-language guidance, and then seamlessly connect with a real dermatologist when needed.

What it does

AcneScan is a web app that:

Lets users upload a face photo.
Uses a computer vision model to detect common acne types and estimate severity.
Summarizes the scan into a simple, friendly explanation.
Uses a Machine Learning model to generate general, non-diagnostic skincare suggestions tailored to the acne type and severity.
Lets users create an account and save their scans over time.
Connects users with board-certified dermatologists, who can view past scan reports and accept appointment bookings.

We are very intentional about safety: AcneScan does not provide a medical diagnosis, prescribe treatment, or recommend specific products. Instead, it’s an educational triage tool that nudges people to seek proper dermatology care.

How we built it

Collected and preprocessed acne image datasets, then trained a TensorFlow CNN to classify acne types / severity.
Wrapped the model in a Flask API (/api/predict) and deployed it on Render.
Built the front end with Next.js, React, and Tailwind CSS, deployed on Vercel.
Used Supabase for authentication, database, and storage of scan metadata (not raw images in the browser) so users can track their history.
Called a pre-trained model via API to turn our outputs into human-readable explanations and general skincare suggestions.
Added a dermatologist directory + booking flow where derms can be discovered by location and users can schedule an appointment.

Challenges

Getting a usable model working on a tight timeline with limited training data.
Making sure we framed everything responsibly from a healthcare perspective, avoiding diagnosis language, adding clear disclaimers, and keeping recommendations very general.
Juggling multi-service deployment: Vercel for the Next.js front end and Render for the Flask + TensorFlow backend.

What we learned

How to take a computer vision model from a notebook into a real, production-ish web app.
Practical trade-offs between model accuracy, inference time, and hosting limits.
How important UX and wording are in healthcare tools—being precise about what the tool doesn’t do is just as important as what it does.

What’s next

Improve the model with better, more diverse training data and clinical input from dermatologists.
Add stronger privacy controls and clearer consent for image handling.
Build a lightweight dermatologist dashboard to review patient scans, see trends over time, and manage bookings.
Explore adapting the tool for other skin conditions while keeping a strong “education, not diagnosis” stance.