BilliBaby: Non-Invasive Neonatal Jaundice Detection via Optimized Lightweight Deep Edge AI

Project Summary: BilliBaby is an offline Android mobile application that utilizes an optimized, lightweight Convolutional Neural Network (CNN) to detect neonatal jaundice non-invasively by analyzing real-time images or videos of a newborn’s skin tone and face.
Demo Context: In our live demonstration, we will showcase a fully functional working system executing machine learning inference locally on a smartphone without needing an internet connection. We will demonstrate the complete workflow, including the user authentication flow, live camera/gallery image parsing, and the post-inference display of jaundice probability percentages. Additionally, we will present the underlying architecture of the custom classification head paired with a compressed MobileNetV2 backbone.