At Samsung R&D Institute Bangladesh, I co-developed the end-to-end ML pipeline for Lung Assists, an AI-powered lung-ultrasound analysis feature for Samsung POCUS systems. I evaluated and customized model architectures, with RTMDet-Ins-tiny reaching a test F1 of 0.65 and ICC of 0.77.
I built ONNX and OpenVINO inference pipelines for the feature, reaching 20+ FPS on Intel Core Ultra NPU through model and runtime optimization. I also established unit and regression testing for model and inference-engine updates.
For Galaxy Connect, I implemented mDNS device discovery and proposed an optimization that reduced mDNS traffic by 53%. I used WinUI 3 and .NET 8 to develop the desktop app, and collaborated on-site at Samsung Digital City in Suwon to co-develop and deploy it.
On Samsung's Continuity Service, I maintained and enhanced the framework, co-invented a BLE advertisement filtering optimization that reduced power consumption by 14.8%, and developed camera continuity for Microsoft Office products and SamsungNotes. I also developed FamilyMDE and AnnoSpread, an AI-assisted medical data annotation tool using MedDINOv3 and Qdrant.
