Entry level, and not hiding it

The qualification is a Level 6 Diploma in Artificial Intelligence Operations from Al Nafi International Colleges, finished in 2026 with a 90% mark at the oral defence. The word that matters in that title is Operations. It is not a course about model architectures; it is a course about what has to exist around a model for it to survive contact with real infrastructure — and that is the part I have ended up caring about.

Underneath it are ninety-five lab exercises committed one at a time across three repositories: fifty-three on DevOps and CI/CD, thirty-four on security operations and forensics, eight on Red Hat Linux. Each one keeps the commands that were actually run. That is where the reflexes come from — why a health endpoint gets shaped like a liveness probe without being asked, why a scheduled workflow gets a concurrency group, why a detector ships with something that deliberately makes it fire.

Certifications

• Diploma in Artificial Intelligence & Operations, RQF Level 6 — Al Nafi International Colleges
• Machine Learning Certification — Al Nafi International Colleges
• Networking Essentials — Cisco Networking Academy
• Introduction to Cybersecurity — Cisco Networking Academy
• Intermediate, Computer Science — IMCB F-8/4, Islamabad

You do not have to take any of this on trust — all four points below are things you can check by opening a file.

• Negative results get written down. When measuring the hybrid retriever showed the dense half was making rankings worse, that became a documented constant with a threshold and an explanation, not a quietly deleted branch.
• Limitations sit in the README, not in a footnote. Which parsers were verified against live responses and which were written from documentation. Which numbers came from a fallback embedder. Which knowledge base is synthetic. Where mock-mode output should not be believed.
• Everything runs without a key. Three of the four projects stand up end to end with no API key and no network, because a project nobody can run is a project nobody will read.
• Scope boundaries are drawn on purpose. The incident copilot advises and never remediates. The pair engineer generates tests and never executes them. The job funnel leaves LinkedIn alone. In each case the repository says why.

The habit underneath all of that is simple and slightly unfashionable: I would rather publish a system with a known flaw described precisely than a system whose flaws are only discovered by whoever runs it next.

The professional work is the operational half of machine learning. The rest of the time I build things because I want them to exist. There is a fighting game in progress with a roster of more than seventy characters, which is an unreasonable number and is the point. There is an automated operating system, which is early and probably too ambitious, and that is also the point.

Older evidence of the same instinct is still on GitHub: a Crossy Roads clone in three.js, built from the memory of playing the original at nine and described in its own README as horribly optimised; a music visualiser in plain HTML and JavaScript, where the README notes that React was considered and then deliberately refused. Neither is impressive engineering. Both got finished, which at the time was the harder part.

My GitHub status has said 🥨 funny for a while now. I am not going to defend it.

AI engineering or AIOps roles. Remote, or relocation to the UAE, Saudi Arabia or Qatar. Entry level — I would rather be the person on a team who ships the drift detector and the runbook than the person who claims five years of it.

Currently going deeper on LangGraph, specifically human-in-the-loop workflows, and on a LoRA fine-tuning project in PyTorch to learn the training side as properly as I have learned the serving side.

awonaziz786@gmail.com is the fastest way to reach me.