About
I care about automation that makes business sense, not automation that demos well.
I started where most frontend developers start: turning designs into interfaces. What changed things was discovering that the interface is usually the smallest part of the problem. Behind almost every screen someone is retyping data that a machine could have read.
So I went looking for the layer underneath. n8n gave me a way to connect language models to real systems — documents in, validated structured data out, straight into the tools a business already uses. Arabic is where this gets interesting, because most global tools quietly do not handle it, and the region is moving fast toward mandatory digital invoicing.
I still write the frontend. I lead it, in fact — I set the standards, review the code, and mentor the developers. The difference is that I now design the whole path, from the moment a document arrives to the moment someone approves it on screen.
The other half of the job is coordination. I gather requirements directly from clients, define scope and acceptance criteria, break work into tasks, and keep delivery on schedule across frontend, backend, and design. It is less glamorous than the AI part and it is often what decides whether a project ships.
The worst bug I have shipped was not in anyone's code. Two systems each behaved perfectly and disagreed about what a payload meant, so data deleted itself silently — no error, no log, nothing broken. Now every shared boundary gets a written agreement first: the shape of the data, the required fields, the strange cases, and who owns what. It lets two people build in parallel and still fit together on the first try.
A workflow that saves nothing is better than one that saves the wrong thing. My extraction pipelines refuse to persist anything until the essential fields are present and plausible — and when something is missing, they say so and explain why. A confident wrong answer is worse than a slow right one.
My first workflow sent every incoming file to a vision model. But a PDF usually carries extractable text, and a text model reads it for a fraction of the price. One branching decision cut the per-document cost by three quarters — with a fallback to vision when extraction comes back empty. The branch is free; the saving compounds every month.
An automation nobody trusts gets bypassed. A dashboard nobody understands gets exported to a spreadsheet. So I care about the interface as much as the pipeline: clear states, honest errors, and no dead ends.
Anime, chess, badminton, and films. Chess taught me to calculate fast, stay calm, and convert an advantage — which is roughly what a deadline asks of you. I competed in ICPC/ACPC in 2021, and that habit of thinking a few moves ahead never really left.
He is an AI Automation Engineer and Senior Frontend Developer based in Al Khobar, Saudi Arabia. He designs and operates production n8n workflows that connect language models to real systems, and builds the React and TypeScript interfaces on top of them. He also works as a Software Engineering Supervisor, leading frontend development and owning delivery end to end.
Smart Scanner is an Arabic-first AI invoice management product. An employee photographs a receipt; an n8n pipeline verifies it is a genuine invoice, extracts every field with GPT Vision in Arabic and English, and classifies each expense. Accountants then review, approve or reject, comment, and export through role-based dashboards. Nour built the entire AI automation engine and the complete frontend.
Arabic is his native language and it is a core part of his technical specialisation. He builds Arabic-default bilingual interfaces with full right-to-left support, and his document AI pipelines are tuned specifically for Arabic invoices, including handwritten ones and ZATCA e-invoice fields — a case most global tools handle poorly or not at all.
Yes. He is open to opportunities in Saudi Arabia and remotely, on-site, hybrid, or fully remote, and takes on freelance automation and frontend projects. The fastest way to reach him is email.
n8n for orchestration; the OpenAI API including GPT-4o and GPT-4.1 Vision with structured JSON outputs; AI agents with custom tool-calling; webhooks and REST integrations; Google Sheets, Google Drive, MySQL, and Telegram as data and delivery layers; and the Model Context Protocol for connecting models to real tools.
By email at nouraldeentofi@gmail.com, or through LinkedIn. This site also publishes a machine-readable profile and an MCP server, so an AI assistant can look up his experience and projects directly.