Alex283h Luzgin

I’m a Computer Vision and Machine Learning Research Engineer with 10+ years of experience building visual recognition systems and deep learning models. My work focuses on representation learning, metric learning, and self-supervised learning, from neural architecture experimentation to production-ready deployment.

I architected and independently developed the ML/CV component of a multi-camera visual recognition system for automated product identification and user tracking. Using a modified EfficientNetV2 for synchronized multi-view processing, I delivered 99.98% product recognition accuracy on an internal test dataset and improved robustness across viewpoints.

On the person re-identification side, I built a multi-camera tracking pipeline covering human detection, face detection, face embeddings, person re-identification, and track matching with the Hungarian algorithm. For ReID training, I applied self-supervised pretraining using synthetic person datasets (e.g., the ClonedPerson dataset) and approaches including VICReg, NNCLR, and SimSiam.

I also develop research methods and deployment pipelines end-to-end: comparing metric learning losses (ArcFace, CosFace, CircleLoss, CenterLoss), running self-supervised learning experiments, and improving embedding quality. For edge deployment, I convert TensorFlow models to ONNX and TensorRT and deploy to NVIDIA Jetson (Nano, Xavier, AGX) and Google Coral Edge TPU.