MOHAMED NAAJIM

AI-Powered Weld Defect Detection (Ongoing Project)

• Developing a YOLOv9-based system to automate detection of welding defects (cracks, porosity) in marine pipelines, aiming to enhance inspection efficiency.

• Building an end-to-end pipeline—from dataset curation (1,000+ annotated weld images) to model training—with plans for Flask API integration.

• Exploring real-time deployment solutions for industrial environments, focusing on scalability and edge-device compatibility.

Marine Repair & Insulation Document Processing Automation

• Developed OCR solution to auto-extract key details from shipping repair forms, insulation work orders, and delivery documents, eliminating manual data entry

• Created centralized digital repository for marine service records with full-text search capability

• Integrated with company ERP to enable real-time updates for inventory tracking and job costing

• Led the POC development of an OCR model for insurance documents, strategically integrating EasyOCR and Tesseract for optimal text extraction.
• Applied advanced techniques and comparative analysis of EasyOCR and Tesseract to significantly enhance the quality and accuracy of extracted text from insurance documents.
• Implemented rigorous data cleaning processes to ensure high data quality, accuracy, and consistency of the OCR output from insurance documents.

• Developed deep learning models (UNet, DeepLab) for detecting building shadows in aerial imagery, including dataset collection and preprocessing.
• Implemented and compared multiple architectures (FCNN, SegNet) for semantic segmentation tasks, optimizing for accuracy.
• Conducted performance evaluation using segmentation metrics and analyzed model strengths/weaknesses for different scenarios.