At RFID Egypt, I develop computer vision models for image and video analysis, including object detection, tracking, segmentation, classification, and OCR.
I optimize models for real-time GPU and accelerated CPU inference using TensorRT, ONNX, and OpenVINO. I also design multi-camera inference pipelines with NVIDIA DeepStream and serve models using FastAPI and Docker.
For a retail computer vision analytics project, I developed modules for people counting, customer demographics, movement heatmaps, and queue measurement. I independently implemented a Hijab/Niqab classification model by replicating a peer-reviewed paper's methodology, achieving over 97% classification accuracy.
I designed a bilingual Arabic-English RAG pipeline for an asset management system, processing Word documents with technical content and embedded screenshots. My work has also included an asset-tag OCR pipeline with image enhancement, barcode decoding, and automated image-quality assessment.
