Ziad Ayman

At Orange Egypt, I design and operate scalable inference pipelines for production ML, LLM, and computer-vision models using Docker and Kubernetes. I built CI/CD workflows that reduced estimated rollback incidents by 40% and shortened rollout time by 50%.

I build agentic AI systems that connect models to real developer workflows. My work includes an Airflow failure-debugging agent, a self-hosted vLLM-backed internal LLM platform, MCP servers for autonomous branch and PR workflows, and a React/TypeScript GenAI extension for Metabase.

I also engineer data and ML pipelines, from parallel WhisperX Arabic transcription and LLM sentiment analysis to churn and recharge prediction monitoring with MLflow and Evidently AI. I optimize local LLM serving through GPTQ, AWQ, and KV caching to improve GPU throughput and latency.

Previously, at the German University in Cairo, I researched vehicle detection, multi-object tracking, and trajectory prediction using LiDAR and camera data. I developed deep-learning models with PyTorch, including Faster R-CNN, YOLO, and multimodal fusion architectures, while mentoring undergraduate students in scalable applications labs.