SkinSage – AI Skin Condition Screening App

A mobile application that uses computer vision to screen images of skin conditions across 15 categories and surface likely matches for faster, more informed consultations. Rather than diagnosing, SkinSage acts as a decision-support and triage tool — narrowing the gap between "I noticed something" and "a professional looked at it."
The project's foundation was data quality. We started from a raw pool of ~50,000 images and deliberately curated it down to a high-quality ~10,000-image dataset, using a mix of manual and automated masking to isolate lesions and strip out noise before any training began. On top of this dataset we trained and benchmarked 30+ models spanning multiple architectures — CNNs, ANNs, and classical ML baselines — then combined the strongest performers into an ensemble that reached 81.7% accuracy across all 15 conditions, a clear lift over any single model in the stack.
The system was built end to end: TensorFlow for training and fine-tuning the model ensemble, OpenCV for image preprocessing and masking, a Python and FastAPI backend serving predictions, PostgreSQL for the data layer, and a React Native front end for on-device capture and results — cutting patient wait times by roughly 40% through real-time, in-app screening.

Shorter card version (portfolio grid tile):
An AI-powered mobile app that screens 15 skin conditions from a photo. We curated a 50k-image pool down to a clean 10k set and benchmarked 30+ models, combining the best into an ensemble reaching 81.7% accuracy. Built with TensorFlow, FastAPI, React Native, and PostgreSQL.