Deepskin - AI based Skin Disease Detection

Skin diseases affect over 1.9 billion people globally, yet timely diagnosis remains inaccessible due to dermatologist shortages and geographic barriers, particularly in developing nations. This paper presents DeepSkin, an AI-powered Android application for real-time skin disease detection directly on mobile devices. The system integrates EfficientNet-B4, a state-of-the-art convolutional neural network optimized for PyTorch Mobile, with the Groq LLM API (LLaMA 3.3 70B) for conversational AI-driven clinical insights. Built using Kotlin and Jetpack Compose, with Firebase cloud integration and Room database for offline persistence, DeepSkin classifies 23 distinct skin conditions from camera or gallery images, achieving a weighted average accuracy of 90.3% across 8,033 test samples. The system demonstrates superior inference speed (average 312 ms on-device), robust accuracy comparable to server-based solutions, and enhanced user experience through LLM-generated clinical reports. This work contributes a complete end-to-end mobile AI pipeline for clinical-grade dermatological screening and demonstrates the viability of deploying advanced deep learning on resource-constrained platforms.