Founder + AI engineer / Damascus
Two live products in Syria, twelve systems built end to end, and every claim on this page has a link or a test count behind it. Engineering student in AI and cybersecurity, open to remote roles in AI, data, automation, growth and content.
Live products
Open them. They are real software, not mockups.
A private tutoring platform for students across Syria. The public site is live, with teacher-review, booking, commission, and admin flows built behind it. Public booking and payment availability are separate release steps.
A Damascus technology-supply business, online. Full Arabic right-to-left catalog, a product inquiry flow, and a storefront the owner uses every day.
Selected work
Each one has a status. The ones with tests say how many pass. What is still missing is written down too.
Routes engineering tasks across Claude Code, OpenCode and Codex. Capability-tier routing, deterministic fallback, durable task state. A tool I use daily, not a demo.
Read the case study
Ported a typed-decision AI model off Apple-only Core ML onto ONNX Runtime and DirectML, so it runs on any Windows GPU. No text generation, no hallucination risk, calibrated probabilities in one forward pass. First-class Arabic evaluation suite, READMEs in 10 languages.
See the repo
A master and worker control loop. Claude plans and routes, a local Ollama fleet executes, and a GPU mutex serialises the hardware. Tier selection is auditable.
See the repo
Deterministic detection for brute force, impossible travel and port scans. Rules run first, a local LLM writes the incident summary second, never the other way round.
Read the case study
A working Claude Code environment: 274 skills, 165 agents and a 10,000-item offline capability index that stays off-context until searched. Routing, hooks and a security build gate wired in.
See the repo
Local-LLM contract extraction that reports missing fields honestly instead of guessing.
Read the case study
Answers business questions only from executed SQL results. Hallucination-guarded by construction, not just by prompting.
Read the case study
Universal Chrome control for any AI agent. Raw CDP command line, no extension, watchdog-bounded verbs. Works with OpenCode, Codex and Claude.
Open the project
Turns long videos into short vertical clips. An attention-ranking engine, GPU transcription with an NVENC fallback, multi-speaker cutting. FastAPI and Next.js.
Ask for a walkthrough
A Claude Code skill and installable plugin. Finds, judges and downloads real design and motion references, cuts out subjects, and wires them into a frontend build.
See the repo
An executive dashboard on top of a real CSV validation, cleaning and transform pipeline. Not static chart mockups.
Read the case study
A free-first B2B sales-intelligence engine. Buying-signal detection, ICP company discovery, transparent scoring and outreach drafting, with zero paid APIs.
See the repoHow I work
Local LLM deployment with Ollama, RAG pipelines, tool-calling agents, hallucination-guard design.
Next.js, React, TypeScript, Arabic RTL storefronts, auth, bookings, deployment on Vercel.
ETL pipelines, validation and cleaning, SQL and relational modeling, Python, browser automation.
Threat-detection logic, secret scanning, secure config, Git, CI-style verification, Docker.
About
Zuhair Alwazzour (زهير الوزّور) is a Syrian entrepreneur and athlete based in Damascus, and the solo founder of Ostazi, a private tutoring platform serving students across Syria.
I build the whole thing myself: the database, the backend, the interface, the deployment and the support inbox. That is why the projects above come with a status. I know what breaks after launch because I am the one who gets the message.
If you have an idea, or a role, the fastest way to reach me is below.
Contact
Direct email: zoherwazz6@gmail.com
P.S. Not sure yet? Read the Open Axis case study. It shows how I work: the tests, the trade-offs, and what is still not done.
P.S. One sentence about what you want built is enough to start. I will reply with the first questions.
Social, carousels and motion
The same discipline applied to content, not just code.
Real client carousels, scored by hand against a fixed craft rubric before anything ships. Same rubric the production QA gate checks for.
Rubric: hierarchy (can you tell what to read first), legibility, deliberate color use, professional polish. Reviewed slide by slide, not auto-generated praise. I'd rather show a 6.9 than fake a 10.
What this proves
Carousels for Alwazour Technical, Alwazour Studio and Ostazi go through the same pipeline as the code above: real reference research, a written hook, a beat map, then a render, checked against a rubric before it ships, not eyeballed once and posted.
See the real accounts
These aren't mockups sitting in a folder. Alwazour Technical's Instagram runs on this exact content system.
Open @alwazour.tech on Instagram ↗