Ali A. Samir

• Delivered flawless release cycles for two major UAE enterprise clients, achieving 100% system reliability and accelerating scalability by driving the technical roadmap, integrating critical features, and enforcing rigorous QA standards.

• Spearheaded the architectural redesign of a Smart City platform, rebuilding the end-to-end AI pipeline to process RTSP video streams, resulting in a 20% increase in system reliability and improved fault tolerance in production. • Orchestrated the integration and calibration of serval AI Models, driving a 10% increase in precision and a 35% reduction in false positives through targeted PyTorch retraining, threshold tuning, and CVAT data validation. • Aligned AI alert probability thresholds with operational SLAs, improving alert reliability and boosting operator confidence in automated detections. • Accelerated multi-model inference throughput by 127% (36 to 82 FPS) by re-architecting the inference engine using optimized matrix operations, software design patterns, and hardware-accelerated GPU execution. • Led the optimization of production License Plate Detector, achieving 97.8% mAP@50 and 97% mAP@50-95 by engineering a YOLOv11-based Polygon detector and implementing a custom data augmentation pipeline. • Spearheaded technical training for 7 engineers, implementing standardized MLOps workflows for data versioning and storage that optimized department-wide data management and accelerated model training cycles.

• Rescued a failing PoC (28% to 89% accuracy) by leveraging strong stakeholder communication to uncover hidden integration constraints, leading to a successful architectural redesign that synchronized the AI and Embedded subsystems. • Slashed OCR inference time by 65% while achieving 99.84% text accuracy (0.8% CER) by architecting a custom single-shot text recognition model with ConvNeXt blocks. • Overcame critical dataset imbalance to enhance model generalization, engineering a synthetic data generator that leverages generative AI and advanced image processing to produce diverse training samples. • Boosted production detection rates by fine-tuning a custom vehicle detector, achieving 83.5% mAP@50 and outperforming the baseline pretrained model by 20.8%

Engineered foundational data management and processing pipelines, implementing robust versioning and storage protocols to ensure data integrity for large-scale AI training. Trained generative models to synthesize data for imbalanced datasets, effectively solving critical data scarcity issues and improving downstream model performance. Played a key role in team expansion by onboarding new hires to the company infrastructure and establishing best practices for data handling and early-stage development.