Selected projects
Things I've built.
A short selection — wearables and medical imaging, tabular ML, and multi-modal CV/NLP — chosen because each one taught me something about shipping models past the notebook.
-
Wearable · ML · Android
HealthConnect
End-to-end health monitoring: Samsung Galaxy Watch (Wear OS) → Android app → FastAPI backend with ML-based ECG classification. Collects 8 biometric signals and runs 1D-CNN, XGBoost, and SVM models for 5-class arrhythmia detection on live ECG.
-
Computer Vision · Medical
MedVision AI
Breast ultrasound analysis with an EfficientNet-B0 3-class classifier and a U-Net + ResNet-34 segmentation head. Full-stack web app with FastAPI backend and a React frontend featuring drag-and-drop upload and tumor overlays.
-
Tabular ML · Finance
Credit Default Prediction
Merged 6 financial data sources into a unified 90K-record, 85-feature dataset. Engineered 27 domain features and built a LightGBM pipeline with class-weighted training for imbalanced default prediction.
-
Computer Vision · NLP
Emotion Detection System
Analyzes human emotion from images, video streams, and text using CV and NLP. Recognizes happiness, sadness, anger, surprise, and neutrality across modalities.