At Club Duelz, I fine-tuned a unified YOLOv11m detector for players, balls, goal posts, and referees, achieving 96.9% mAP50 and 95.4% precision. It replaced a two-model pipeline and halved inference passes per frame.
I built a real-time computer vision pipeline using YOLOv11m, SigLIP, and BoT-SORT. It’s live for the AIFF National Futsal Championship 25–26 season, powering possession, pass, and interception stats.
At ODDeven, I translated an AAG-based GNN research pipeline into a production system for manufacturing quotes, reducing quote-generation time by 30%. I also developed a hybrid feature recognition method using AAG representations and XGBoost.
At TechXi, I built a real-time hand gesture recognition system using OpenCV and MediaPipe, and trained TensorFlow models to convert signs into speech. I’ve also contributed to Club Duelz’s Python API and platform, which has reached 10,000+ registered users across 20+ tournaments and 3+ state football associations.
