About Me

I am a Vision AI engineer focused on bringing deep learning-based object detection models into real-world environments.

I work primarily with YOLO-based models, improving performance and optimizing them for practical use. I build real-time perception pipelines using DeepStream and tracking, and design camera-based systems that operate reliably in low-latency streaming environments (e.g., RTSP).

Rather than focusing only on model accuracy, I approach problems from a system perspective—considering how models behave and perform in real deployment scenarios. I design and implement the full workflow from training and inference to optimization and real-time execution.

In addition, I have experience with time-series modeling (e.g., BiGRU, ConvLSTM), data analysis and visualization, and building components that connect models to the real world, such as pixel-to-GPS transformation systems.

This blog documents my process of building Vision AI systems, including experiments, implementation details, and lessons learned along the way.

Skills & Technologies

• Deep Learning: PyTorch for object detection and time-series modeling
• Computer Vision: OpenCV, YOLO-based detection, tracking
• Edge AI & Deployment: TensorRT, ONNX, NVIDIA Jetson, DeepStream
• Systems & Optimization: real-time video processing, performance optimization, C++ modules
• Data & Analysis: time-series data analysis, Plotly-based visualization
• Development: Python, C++