Fine-tuned an InceptionV3 CNN on Food-101 with mixed precision, custom LR schedule, and checkpointing on an RTX 3090, reaching ~80% validation accuracy.
Containerized a personal media ecosystem for reliable remote access, metadata retrieval, and easy maintenance using Docker Compose on Ubuntu.
Designed a deep-learning pipeline with a custom 3D CNN and a motion-detection branch using temporal convolutions for event detection in baseball videos.
Implemented a Double DQN agent with a custom Gymnasium environment to learn partial playthrough of the first Geometry Dash level.
Built an Android app using MVC with Firebase for remote storage and SharedPreferences for local state. Added rating UI, custom dialogs, and responsive layouts.