At Ménara Préfa, I developed and evaluated a five-step pipeline to detect, segment, and classify concrete surface defects. The models included YOLOv8, U-Net, and ResNet-18.
I automated the image-to-diagnosis-to-report quality-control process and deployed it as an inference service using Gradio and Hugging Face Spaces. I also designed severity indicators and inspection reports for quality teams.
In a personal project, I developed a URL fraud detection service using XGBoost, FastAPI, and Docker, with a React monitoring dashboard. At AIKINI Web & Tech, I designed and deployed an e-commerce platform, and my academic NLP project compared TF-IDF with LSTM for sentiment analysis.

