Masoud Masoumi Moghadam Email & Phone Number

Türkiye

Amsterdam, North Holland, Netherlands

• Deployed and served model to detect volleyball actions in real-time and save the extracted KPIs on a timestamp-based database.
• Trained and Tested multiple deep learning models to detect, extract, and recognize the P&ID and PFD symbols which contain 400 different objects.

Singapore

• Developed and deployed an AI model to detect cricket deliveries in live-stream matches, achieving a 95% accuracy rate, which improved the real-time analysis of match events.
• Developed AI-integrated solutions for automated KPI extraction, enabling 80% faster insights into player performance and ball trajectory, improving decision-making for teams and analysts.
• Optimized inference pipelines, achieving 2x improvement in model speed performance by employing techniques such as model pruning, quantization, and leveraging ONNX, TensorRT, and TensorFlow Lite for reduced latency and.
• Integrated heuristic algorithms on deep learning outputs, leading to a more sophisticated and accurate analytical results for player performance metrics, enabling more reliable performance evaluations and strategic.

Tehran, Iran

• Contributed to the development and optimization of the Signal app’s remote authentication service, reducing video processing latency by 6 seconds per client, resulting in a 20% improvement in user experience and system.
• Enhanced the precision of a real-time object tracking system in an object re-identification project, improving accuracy by 7% through ONNX and TensorRT optimizations, boosting system reliability and effectiveness in.
• Led and supervised a 3-person team to deploy a face feature matching solution on edge devices using a one-shot learning (Siamese) network, delivering a successful deployment within 20% less time than initially.

Iran, Mashhad

• Deployed an object segmentation model for wind turbine blade inspection, successfully isolating blades from complex backgrounds and stitching segmented images into a panorama with 98% precision. This solution.
• Developed and implemented an attention-based deep learning model for detecting illegal constructions in aerial images, achieving 94% accuracy in identifying unauthorized developments and automating reporting, reducing.

Hamburg, Germany

• Implemented an optimized YOLO model for face and mask detection, resulting in a 30% improvement in precision-recall curve performance over the prototype model, significantly increasing detection accuracy and reliability.
• Enhanced production inference speed of the face/mask detection model from 4 FPS to 10 FPS by applying model quantization and pruning techniques, achieving a 150% increase in processing efficiency, which reduced latency.

Khorasan-e Razavi Province, Iran

• Developed a real-time traffic prediction model leveraging historical traffic data, weather conditions, and event-based factors, resulting in a 15-20% improvement in route optimization and ETA accuracy.
• Implemented AI-driven personalized routing that adjusts routes based on user behavior and preferences (e.g., fastest route, scenic route, mode of transport), increasing user engagement and satisfaction by 10%.
• Designed an anomaly detection system for identifying unusual traffic patterns, such as accidents and roadblocks, enabling real-time rerouting and reducing user delays by 20%.