Achraf LAMIA

•Designed a real-time computer vision system to monitor textile production lines, track finished pieces per operator/station, and provide a data-driven basis for performance steering. •Developed a multi-object detection and tracking pipeline (DF-DETR + DeepOcSort) enabling automatic counting and precise cycle-time measurement per piece and per worker, in order to identify and eliminate bottlenecks. •Implemented a gesture recognition module (YOLO + BiLSTM) to detect incorrect or risky textile gestures, suggesting corrections to reduce micro-losses in productivity and protect operators’ health (posture, repetitive movements). •Optimized models for real-time inference (ONNX, TensorRT, INT8 quantization) and continuously streamed KPIs (TRS/OEE, productivity, risky-gesture alerts) to a supervision dashboard, with the goal of bringing per-piece manufacturing times closer to leading international textile standards.

• Designed, trained, and deployed two YOLOv8 models: ° Player/goalkeeper/referee/ball detection (European dataset + fine-tuning on African match frames) for robust, accurate results across continents. ° YOLOv8 Pose for advanced pitch keypoint detection, powering a radar view with a custom homography matrix. • Developed a proprietary dynamic homography adjustment technique, enabling reliable transformation even with camera zooms or missing keypoints—addressing real-world challenges in African stadiums (variable camera angles, lighting, conditions). • Optimized multi-object video tracking (StrongSort, ViT-B-16 ReID) and fully automated player team classification (SigLIP, UMAP, KMeans). • Delivered advanced tactical analysis: automatic formation detection, heatmaps, spatial mapping, control zone and open space identification, real-time player/ball speed calculation. • Deployed a robust AWS cloud pipeline: auto-managed EC2 (GPU/CPU), S3 (Standard/Glacier), Load Balancer for scaling and multi-user API access, CI/CD automation (Docker, Jenkins), and secure IAM practices. • Presented and validated the system to AI experts, sports professionals, and investors—demonstrating real-world impact and scalability.

• Development of a fine-tuned LLM model using LLaMA 3 to automate the identification of the most suitable vacant positions for employees in surplus. • Implementation of a data pipeline to manage job descriptions and employee resumes. • Development of an application in Microsoft Access, integrating SQL and VBA, to ensure efficient automation and secure centralization of HR data. • Development of a Flask API to connect the database to the LLM model for fine-tuning. • Creation of a dashboard with Power BI for dynamic visualization of KPIs to support strategic decision-making. Department: HR IT Team Sector: Banking

• Implementation of a segmentation system based on current consumption levels using the K-means clustering method. • Prediction of customers' future consumption using the Support Vector Machine (SVM) method. • Development of classification models to estimate the probability of customer payment default using the Naive Bayes method. Department: Information Systems Sector: Energy Distribution

• Artificial Intelligence, Computer Vision, Big Data, Deep Learning.

•Areas of Expertise: Statistical Physics, Analysis, Algebra, Digital Electronics, Computer Science (C and C++).