At LEN, I built an AWS video pipeline from scratch for race shoe counting, processing races with up to 20,000 runners. I developed detection, pose, tracking, segmentation, and shoe-identification systems, reaching about 98% brand precision and 95–98% shoe detection accuracy.
At ShotQuality, I designed a shot-detection model that improved precision by 25% and deployed it on AWS SageMaker. I also unified scoreboard detectors and replaced RNN-based OCR with a Qwen VLM, improving accuracy by 10%.
At Atos & Eviden, I worked on computer vision for a smart-city program running models on more than 30,000 real-time cameras. I also built an automated model retraining and deployment tool and led four junior engineers on a retail analytics project.

