Mosbeh Barhoumi

Developed a sophisticated deep learning system for car parts detection and damage estimation using YOLO, Detectron2, mmdetection, and various other architectures and CNNs, integrating fraud detection capabilities. Created a system for car damage repair cost estimation. Developed a semi-supervised multi-class damage segmentation pipeline using YOLO. Built a machine learning solution for text extraction from invoices, including table reconstruction, and VIN and license plate extraction using PaddleOCR, Tesseract, and various other OCR engines. Evaluated the efficiency of AI solutions in terms of resource consumption using metrics and libraries such as Apache JMeter, Locust, k6, and wrk. Mastered semi-annotation and labeling tools like Label Studio and Roboflow. Developed a car diagnostic system to determine if a vehicle has undergone a comprehensive inspection. Implemented test automation scripts for several projects. Developed clustering solutions to detect anomalies in Automobile Incident Reports. Implemented several APIs using Flask. Deployed all machine learning solutions in production using Docker and Docker Swarm, ensuring the scalability of the systems.

Virtual Try-On Using Deep Learning This project represents an ongoing research and development endeavor aimed at advancing the field of virtual try-on technology. It involves a comprehensive review of over 20 research papers to gain an in-depth understanding of existing methodologies and approaches. Subsequently, a bespoke model architecture is being developed to push the boundaries of virtual try-on applications. Notably, our efforts have led to the establishment of an official pre-processing method for one of the most widely utilized datasets in this domain, VITON.

Multi-lingual Chatbot Supporting Arabic, French, and English In the course of this project, I undertook the development of a multi-lingual chatbot designed to provide support for Arabic, French, and English languages. This chatbot was conceived to serve as a sophisticated solution for the renowned educational platform, TakiAcademy, with the ultimate goal of supplanting the need for human assistants.