I hold a Master’s in Information and Communications Engineering from the University of Trento (Italy) and an engineering degree from Sup’Com (Tunisia). I approach every project with a problem-solving mindset grounded in …
I hold a Master’s in Information and Communications Engineering from the University of Trento (Italy) and an engineering degree from Sup’Com (Tunisia). I approach every project with a problem-solving mindset grounded in both theory and hands-on experience; always aiming to craft systems that learn intelligently and act meaningfully.
Development of an AI Solution for Data Collection, Automatic Annotation, and High-Quality Assurance to Enhance Embedded Neural Networks: Developed an open-set unsupervised annotation pipeline by combining segmentation (SAM), embedding (CLIP image encoder), and clustering, enabling easy prompting and handling novel classes.Established a benchmarking system using COCO metrics and dataset subsets to evaluate and optimize models, achieving 80% AP for the person class and 60% AP for the bowl class with tailored filters.Boosted AP by 10% for both classes by combining VIT32 and VIT16 embeddings for enhanced features.Optimized pipeline options for a 7x speed boost with
Developed an open-set unsupervised annotation pipeline by combining segmentation (SAM), embedding (CLIP image encoder), and clustering, enabling easy prompting and handling novel classes.
Established a benchmarking system using COCO metrics and dataset subsets to evaluate and optimize models, achieving 80% AP for the person class and 60% AP for the bowl class with tailored filters.
Boosted AP by 10% for both classes by combining VIT32 and VIT16 embeddings for enhanced features.
Optimized pipeline options for a 7x speed boost with
Conducted an in-depth study of hand gesture recognition models, ensuring a comprehensive understanding of current trends and advancements in the field.
Utilized Mediapipe for hand and landmarks detection, optimizing cursor movements (following hand movements) and gesture recognition with a two-scale motion approach.
Analyzed diverse aerial datasets, implementing Mask RCNN, YOLO, and SAM with box prompts.
Outperformed with YOLO in medium-sized instances data, surpassing other models by 20% in F1 score, 10% in mAP, and speed.
Excelled with Mask R-CNN in small-sized instances, achieving nearly 50% mAP, outperforming YOLO by almost 15% in mAP, with faster inference speed than YOLO+SAM.
نظام التوظيف التشغيلي المعتمد على الذكاء الاصطناعي أولاً، الذي يوحّد ويؤتمت دورة التوظيف بالكامل للمؤسسات الحديثة حول العالم.
د.جوب منطقة حرة ذ.م.م. 2026 © جميع الحقوق محفوظة
