Sami Al-Rabian — AI Engineer

AI Engineer — I design and build AI systems and agents. Genuinely into AI, and a Python developer by strong preference because of how fast it lets you move.

GitHub: github.com/0c33 — Email: [email protected]

Security

I’ve found several vulnerabilities before, mostly on iOS. The most notable were three vulnerabilities in one application, found within a Bug Bounty program organized by the Saudi Federation:

• An authentication bypass on the employee login system.
• An impersonation vulnerability allowing messages to be sent as another employee.
• A full database leak of employee data — including phone numbers, emails, names, and other sensitive information.

I hold the eJPT certification, and have hands-on experience in Reverse Engineering, with a full, integrated toolset environment — currently running as a VM on Proxmox, one of my main environments for this work.

I also have practical experience analyzing application traffic through Burp Suite, manipulating requests and responses on iOS. I run a full iOS 17.0 jailbroken environment, used strictly for penetration testing — without causing any unintended or illegal harm, and only within legal, authorized scope.

Locally Hosted Models

I’m currently hosting 3 AI models concurrently across my infrastructure:

• Qwen3.6-35B-A3B / 27B are the two primary models behind my work and projects, and also the main model powering Hermes Agent. I run them through llama.cpp on the Vulkan backend — I tried ROCm earlier but it was slower, so I settled on Vulkan.
• Ornith 1.0 9B performs excellently as an agentic model, and I currently use it as a secondary model for Hermes Agent.
• Vibethinker 3B I use for experimentation, as a light CPU-only workload.

Each of my servers hosts one model independently — the first runs Ornith 1.0 9B, the second runs Vibethinker 3B.

I’m currently building a fully integrated, AI-dedicated infrastructure. I also host more than one site — one of them is this blog, hosted on one of the servers.

Networking

I work on networking and genuinely enjoy it. My home network currently runs entirely on OPNsense as a VM on one of my Proxmox servers. I have two servers, each with 32GB RAM, and I’m very comfortable with them.

I recently added an Intel X540-T2 card to my internal network — two Ethernet ports, 10 gigabit each. I set up full passthrough to OPNsense, so it has complete, direct control of the card with no extra layer in between — zero latency.

I also passed through an RTX 2060 Super to a Windows 11 VM, mainly so I can stream gaming to every device in the house — using Sunshine as the streaming server and Moonlight as the client to display the stream and play over the internal cloud. The quality is excellent, with no lag even in input/control.

Current Focus

My full focus right now is AI — running it, using it, and building on it.

My Work & Projects

I previously built an AI system that improves itself based on how the client uses it — Agentic AI Architect:

1. Give it the idea, and it starts asking questions until the agent fully understands you and the picture is completely clear.
2. It then builds a prompt, inputs, and the required outputs.
3. It runs the new agent against the new inputs, checks the result, and compares it against the client’s request — is it aligned or not.
4. If the results aren’t suitable from the agent’s own perspective, it retries — and states the reason for the retry.
5. If the agent judges the results satisfactory, it goes and asks the client: is this result satisfactory or not?
   • If satisfactory → moves to the next step.
   • If not → it asks the client what specifically isn’t satisfactory, with details and notes, and repeats the process — until the client agrees. This way the result ends up 100% matched to the client’s taste.
6. Next step: it tries to build a full environment for the new agent, runs it, and resolves any issue it runs into.
7. If it judges the code isn’t excellent, it gives itself feedback and revises it. If it judges both the results and the code satisfactory, it asks the client: are the results satisfactory to you, and is the code excellent?
   • If no → the client gives feedback and the agent works through it, and so on.
   • If yes → it saves the code, names it, and the new agent is ready to use.

This is my flagship agent — and the next one, God willing, will be even better.

Want a service done?

Just reach out, explain your idea and the goal, and give me an example of the required inputs and outputs — and it’ll be worked on at the lowest cost and best results.

📧 [email protected]

My success is Saudi Arabia’s success, and Saudi Arabia’s success is ours.

Sami bin Mohammed bin Sulaiman Al-Rabian SBM Labs KSA 2030