Egypt · Open to remote & relocation (KSA/UAE) · marwanmaher635@gmail.com · LinkedIn · Portfolio
I am the Project Manager for IT projects at an IP-services firm in Riyadh, and I have been building software since 2022. I run the IT portfolio, five production systems, from requirements and vendor evaluation through release and maintenance, in a domain where being wrong has legal consequences. The products are my employer's and stay internal, so this profile shows the part that generalises: the tools I build in the open, and how I work.
| Tool | What it helps you do | Stack | Tests | Link |
|---|---|---|---|---|
| decision-matrix | Make a build-vs-buy or tool choice you can defend: must-haves kept apart from weights, how far a weight must move before another option wins, and a decision record you can commit | Vue 3 · TypeScript | 63 | repo · live demo |
| phasegate | Walk a project from discover to operate, and fail CI when a phase the team has moved past is not actually done | TypeScript · CLI · GitHub Action | 86 | repo |
| django-postmortem | Run blameless incident reviews where every action item names the guard that stops the failure coming back | Python · Django | 133 | repo |
| Tool | What it helps you do | Stack | Tests | Link |
|---|---|---|---|---|
| vue-rtl-kit | Stop dates, phone numbers and order references reversing inside Arabic text, with direction state that survives SSR and logical-property utilities | Vue 3 · Nuxt · TypeScript | 106 | repo |
| django-rtl-admin | Make the Django admin properly right-to-left: logical-property styling, isolated cell values, locale numbering systems and a language switcher | Python · Django | 192 | repo |
In an Arabic sentence, a date or a version string silently reverses, and users report it as "the number is wrong" rather than as a rendering bug. Both kits exist for that.
| Tool | What it helps you do | Stack | Tests | Link |
|---|---|---|---|---|
| claude-account-switcher | Run several Claude Code accounts and fail over when one hits its rate limit, carrying the conversation across | Bash · Python | 232 checks | repo |
All six are MIT-licensed and run their tests in CI. Also: vue-intent, a Vue 3 adapter for intent-first, state-driven UIs with permission guards and navigation protection.
- IT portfolio, as Project Manager: five production systems, from requirements and procurement through release and maintenance.
- IP management platform: built end to end on a commercial admin base (data model, modules, role-scoped access, backend integration, bilingual Arabic/English RTL). 52 IP services live; the daily system of record.
- Trademark monitoring: wrote the requirements, ran a four-vendor evaluation, made the build-vs-buy call, then led the in-house build. In production.
- Quality and delivery: founded the QA practice from nothing (530+ Selenium test cases across 170+ features, manual regression down 40%), wrote 59 pages of developer documentation that halved onboarding time, and led a cloud migration with zero downtime.
- Cepro.ai: Next.js 15 + React 19 rebuild, 17 pages, in-browser Arabic OCR (Tesseract.js), delivered two weeks early.
- siprc.sa: built the company website (TypeScript, Vite, GSAP; Arabic/English) and its AI customer-support assistant.
- Before that: sole frontend engineer at Digitee, shipping six production apps including an AR virtual try-on and a Three.js 3D showroom, and extending a multi-locale setup to 16 languages with full RTL. Earlier, frontend engineer on Keme, a US clinic-management platform, at IDAAM.
Internal products are described at the level of my role only; implementation details belong to my employer.
- Own the portfolio, not the ticket queue. Write the requirements a vendor is judged against, run the evaluation, and be accountable when the answer is to build instead of buy.
- Decide what not to build. Most operational pain is a workflow problem wearing a feature request as a disguise. One finished reporting build was rejected because it missed what the business needed; re-scoping it was the right call, not a setback.
- AI where it is accountable. Model output that drives a legal recommendation is validated against historical outcomes before rollout.
- Look at the real data before writing the parser. More than once, a field I "knew" the shape of turned out to have two shapes, and the difference mattered.
- Test in the real thing. Scripted probes report success on broken paths. A user interface is not done until it has been driven in a browser.
- Make the failure mode expensive to reintroduce. A bug that caused real damage earns a test, a comment explaining why, and a line in the README.
- Say plainly what broke. Post-mortems that hedge teach nobody anything.
Stack: Python · Django · PostgreSQL · Vue 3 · Nuxt · TypeScript · Next.js · React ·
Selenium · GitHub Actions · Arabic/English RTL
Contact: marwanmaher635@gmail.com · LinkedIn · Portfolio