My work sits between model training and usable software. I curate datasets (LabelImg, Roboflow), train YOLO variants on Google Colab and local GPUs, validate with mAP/precision/recall, then wrap results in PyQt5 dashboards or plan Jetson Nano deployment. I care about the full loop — not just the accuracy number.
01
Training pipeline
Dataset annotation, augmentation, hyperparameter tuning, and mAP validation on YOLO architectures.
02
Application layer
PyQt5 GUIs, OpenCV preprocessing, real-time inference loops, and visual reporting.
03
Edge deployment
TensorRT optimization, Jetson Nano constraints (4GB RAM, 128 CUDA cores), and latency profiling.
