Projects

Investigated neural scaling laws on symbolic music by training 9 decoder-only Transformer and LSTM models (844K–201M parameters) on 100M+ tokens of ABC notation from 54K+ folk tunes

Ranked 1st in the Kaggle competition by achieving 96%+ accuracy on US state classification with a 7-model ensemble strategy

A comprehensive object detection system implementing YOLO (You Only Look Once) algorithm combined with DeepSORT tracking for real-time multi-object tracking and detection in video streams. Features include pedestrian detection, vehicle tracking, and customizable object classes with high accuracy and performance.

Implemented semi-supervised learning with the MixMatch algorithm on a custom 500 image dataset.

Web-based system with Django, featuring book search, issue tracking, user authentication, and admin dashboard controls.

This is a frontend website for Military Management operations.