At WADIC, I co-developed and productionized an AI roof analysis platform that brings rooftop imagery, segmentation, pitch estimation, measurement, and reporting into one workflow. It cut average roof assessment time by 87%, from about 15 minutes to about 2 minutes per property.
I trained and refined Mask R-CNN models with PyTorch and Detectron2 on annotated rooftop images. The segmentation system achieved 93% IoU-based performance, while a separate house-boundary model achieved 96% accuracy.
I also developed a roof pitch estimation methodology using mathematical modelling, calibration, and iterative error analysis. I built and deployed the full-stack application with Python and FastAPI, integrating the trained models into a production workflow on AWS EC2.
Earlier, at NOON E-Commerce, I analyzed fulfillment and inventory data and automated campaign reporting. At AL JEMZAWY TECHNOLOGY, I developed a speech-to-text application and a multilingual PDF summarization application supporting English and Arabic.
