MOHAMMAD KHALED MOSLEMANY

• Developed a Python-based pipeline using Facebook SDK to automate the creation of social media poll video ads for surveys in crisis-affected locations. • Fine-tuned Stable Diffusion models for generating video frames and utilized Mistral and Meta LLaMA language models to create ad captions, and poll questions. • Integrated machine learning tools for seamless video generation and captioning, enabling efficient and targeted crisis response through automated social media campaigns.

Built pipelines for video-based job applications, enhancing trust with AI-driven identity and content checks. • Deployed Face Recognition models using DeepFace with various backbones, such as VGG-Face and FaceNet to verify user identity in video submissions. • Integrated AWS Rekognition to detect explicit content, ensuring compliance and content safety.

Fine-tuned SMART and VideoMAE models for emergency action detection (e.g., fainting, fire incidents) and deployed on Nvidia Jetson Nano for real-time processing. • Optimized models using DeepStream SDK and TensorRT for efficient edge performance, connecting to surveillance cameras for continuous monitoring. • Integrated edge devices with Azure Cloud to automate alerts, sending messages or making calls to relevant authorities, such as hospitals or emergency responders, for timely intervention.

• Developed and deployed OCR models using EasyOCR, fine-tuning pipelines for Arabic text extraction from documents such as IDs, driving licenses, and invoices. • Built and optimized the OCR pipeline end-to-end, including preprocessing (noise reduction, binarization, and skew correction), text detection with CRAFT, and recognition using CRNN.