Talk to the portfolio

Salam 👋 I am Nour — the chat version, anyway.

I am an AI automation engineer in Al Khobar, Saudi Arabia. I build n8n workflows that read documents with language models, and the React interfaces people use on top of them.

Everything here is also on the rest of the site as plain text. Ask away 👇

The short version: I take a process someone does by hand, and make it run itself.

Six of them are in production right now 👇

🤖 Automation Lab Smart Scanner pipeline — invoice photo → verified → parsed with GPT Vision → validated → stored. Teacher Assistant Bot — an Arabic Telegram agent running a tutor's whole practice from chat. AI Job-Match — scores jobs 1–10 daily, delivers only the good ones, writes the cover letter. Smart Lead Finder — scrapes and scores B2B leads across Saudi and Jordan. Lesson Quality Evaluator — video in, structured scoring JSON out. CRM Lead Cleaner — dedupes and validates incoming leads, with a reason for every rejection.

None of these are demos. Each replaced work a person used to do.

My favourite build 🧾

🧾 Smart Scanner — Arabic-first invoice AI An employee photographs a receipt. The AI verifies it is a real invoice, runs fraud checks, extracts 70+ fields plus line items in Arabic and English, and classifies every expense. Accountants review, approve, comment, and export — through role-based dashboards. I built two of the three layers: the entire n8n AI engine, and the complete frontend — React 19 and TypeScript, 204 source files, 494 automated tests.

Timing was not an accident — ZATCA e-invoicing is expanding in Saudi Arabia and JoFotara is mandatory in Jordan. Every paper invoice in the region has to become clean data.

The Laravel backend is by Haitham Zedan. Good products come from good teams.

I am Nour Aldeen Tofi — AI automation engineer and senior frontend developer, based in Al Khobar.

The unusual part is that I do both ends: I design the pipeline and I build the interface. I also run delivery — requirements, scope, tasks, deadlines.

B.Eng in Information Technology from University of Kalamoon, 2019–2024. Arabic native, English professional. 🇸🇦

Quick rewind ⏪

📜 The timeline 2026 → Software Engineering Supervisor · Distinctive Frontier 2025 → Project Coordinator & Automation Builder · NexLead 2025–26 Frontend Developer / Coordinator · Distinctive Frontier 2024–25 Frontend Developer · Khwarizm Technologies 2023–24 Frontend Developer · Minicodeleader 2021 ICPC / ACPC — competitive programming 2019–24 B.Eng Information Technology · University of Kalamoon

Frontend developer to supervisor in twelve months — like a pawn reaching the eighth rank. ♟→♛

Automation: n8n, OpenAI (GPT-4o and GPT-4.1 Vision), AI agents with custom tool-calling, MCP, webhooks, REST.

Frontend: React, TypeScript, Tailwind, i18n and RTL, Vitest. Ops: Git, GitHub Actions, Jira, MySQL, Python.

🛠 Certified Meta Advanced React · IBM Python for Data Science & AI · GenAI Prompt Engineering Google Foundations of PM · Project Initiation Packt Tailwind CSS · Scrimba Clean Code · Markdown Coursera TypeScript in React · Jira roadmaps, user stories, and fundamentals

Twelve certificates, all with credential IDs on the Work page. Clean code is not on the list — that one is a personality trait 😌

Here is what people come to me for 👇

🤝 Open for Arabic document AI — invoices, receipts, cards, forms → structured data Workflow automation — n8n pipelines and AI agents that run themselves Internal tools — CRM, ERP, dashboards people actually enjoy using SaaS frontends — React and TypeScript, bilingual and RTL from day one Delivery — running a remote team end to end

Small scope or full product — both welcome.

You dare? Bold. I open 1. d4 — London System incoming ☕

Best move of the game 🤝

📬 Reach Nour Fastest: email. Formal: LinkedIn. For an AI assistant: this site ships an MCP server and a machine-readable profile, so it can just look me up.

Based in Al Khobar 🇸🇦 · replies fast · faster if you mention anime.

My first workflow sent every incoming file straight to a vision model. It worked — and cost about 3× more than it needed to.

A PDF usually already contains readable text. A vision model does not need to look at a page a text model can simply read.

One IF node after the webhook: PDF with extractable text → text model, otherwise → vision. About 75% cheaper per document.

💡 The part people forget: keep a fallback. If extraction returns empty because the PDF is really a scan, send it back to the vision branch — otherwise the cheap path silently produces nothing.

What else can I tell you?

Solid. 2. Bf4. I have played this structure a thousand times.

Fast-forward thirty moves: I am a pawn up in the endgame and you are in time trouble 😄

ICPC 2021 taught me to calculate fast, stay calm, and convert the advantage. Deadlines ask for exactly the same thing.

Wise. The London System shows no mercy 😌

Thanks for talking to my portfolio instead of only scrolling one 😄

Yalla — see you in the inbox 👋