I build machine-learning systems that have to survive contact with real infrastructure.
I'd rather hand you an honest number than an impressive one.
-
AI Engineer
Computer vision and system integration for a large-scale public-safety operations deployment — camera-health monitoring, offline inference, endpoint management, and the dashboards operators actually use.
full case study → -
AI Co-op Intern
Ticket-pricing analytics and a deployed price-prediction tool for the Advanced Data Analytics division, plus Power BI reporting and BRD work that earned the recommendation letter shown here.
full case study →
My projects, strongest first. Each one links through to how it actually went, including the parts that didn't.
-
Camera cleanliness monitoring
Detecting degraded camera feeds across a large deployment, and turning a model score into something an operator can act on. Runs on an offline inference server — a lightweight classifier exported to ONNX and loaded by a single .NET service with no runtime to vendor.
-
Traffic Light Management System Using CNN
Replacing a fixed signal cycle with one that reads how many cars are actually waiting. The interesting number was recall — the baseline detector was missing half the vehicles, and in a counting system that becomes wrong timing.
-
Arabic sentiment classification
AraBERT fine-tuned over roughly 58,000 tweets, where dialect variance mattered more than architecture. The harder part wasn't the model — it was dialect variance and label disagreement that wasn't random.
-
Ticket classifier, containerised
A deliberately small end-to-end service, built to put container fundamentals into practice. It builds and runs — it is not a production system, and it isn't presented as one.
-
Hajj organisation database
Designing the relational backbone for a large seasonal event before writing a line of application code — ER modelling, normalisation, and the stored procedures that enforce the rules a UI can't be trusted to.
-
Image enhancement through morphing
Classic computer vision without a neural net: denoising, alpha blending, and frame-sequence generation to morph one image into another, at a point in the degree before deep learning took over every answer.
I studied Artificial Intelligence at the University of Jeddah and graduated in 2025. Before that I worked in analytics at Saudia, which is where I first got comfortable with the gap between a model that works in a notebook and a system that has to run somewhere real.
Since graduating I've worked where models meet infrastructure — which in practice means Windows Server, restricted networks, and SDK versions that were frozen years ago. That constraint has shaped how I work more than any course did. When you can't reach the internet from the deployment machine, you learn to care about what a dependency actually costs.
In October 2026 I earned NVIDIA's Certified Associate credential in AI Infrastructure and Operations. The exam covers what sits underneath a model: GPU hardware and scaling, networking, power and cooling, cloud versus on-premises deployment, cluster orchestration, and monitoring. It put a formal frame around what the day job had been teaching me from the other direction: the model is the easy part, and the system it runs on decides whether it works. My daily work is that same side of the problem, on Windows Server, restricted and air-gapped networks, and offline deployment. The certification gives me the vocabulary and architecture to take it toward larger AI infrastructure.
In the same month I also passed AWS's Certified AI Practitioner exam. It covers how machine learning and generative AI work, where foundation models fit, and how to use them responsibly, securely and under proper governance. It gives me a shared language with cloud and product teams, and a foundation on AWS to build on alongside my Google Cloud MLOps work.
Outside the day job I've also worked through certifications in containers and Kubernetes with IBM, and in MLOps on Google Cloud. Some of that is now applied, some is still lab work, and I keep the two marked separately rather than blur them.
I write things down. Architecture, decisions, what broke and why. Partly for whoever inherits the system, partly because writing it down is how I find out whether I understood it.
Arabic is my first language. I work in English.
AI & machine learning
PyTorch · TensorFlow · scikit-learn · computer vision · NLP and transformers (AraBERT) · YOLO object detection · ONNX Runtime · model evaluation and metrics · generative AI and foundation-model concepts · responsible AI and governance · AWS AI services (AWS-certified, foundational)
Data & analytics
Python · SQL · Power BI · Power Apps · DAX · data cleaning and transformation · Excel
Cloud & MLOps
Google Cloud · Vertex AI · BigQuery · Docker · Kubernetes · OpenShift · CI/CD concepts — labs and coursework, marked as such under credentials
Backend & web
C# · ASP.NET Core · PHP · Laravel · REST APIs · MySQL · JavaScript, HTML, CSS · Git · GitHub
Systems & infrastructure
Windows Server · Milestone XProtect · Linux and SSH fundamentals · PowerShell · air-gapped and offline deployment · AI infrastructure and operations concepts (NVIDIA-certified)
- BSc Artificial Intelligence — University of Jeddah2025 · degree
- AI Specialist — Saudi Council of Engineers2025 · license
- NVIDIA-Certified Associate: AI Infrastructure and Operations — NVIDIA2026 · certification
- AWS Certified AI Practitioner — Amazon Web Services2026 · certification
- GitHub Foundations — Microsoft2026 · certification
- Cooperative Training Program — Saudia2025 · internship
- MLOps on Google Cloud — Specialization2026 · labs
- Containers, Kubernetes & OpenShift — IBM2026 · applied
- Machine Learning in Production — DeepLearning.AI2026 · applied
- Laravel & PHP — Board Infinity2026 · applied
- AWS Certified AI Practitioner (Learning Path) — Saudi Digital Academy2026 · labs
Marked applied where I have shipped something with it and labs where I haven't. The distinction seems worth keeping.
Based in Jeddah. Open to on-site and hybrid roles in Saudi Arabia.