Projects — Nour Aldeen Tofi

AI SaaS product2026

Smart Scanner

Arabic-first AI invoice management — from photo to approval, with zero manual data entry.

The problem: In most companies an invoice sleeps in a pocket or a drawer until month-end, when a stack of paper lands on the accountant's desk to be typed in by hand. The result is lost time, human error, duplicate and forged invoices slipping through unnoticed, and no record of who submitted what. Global tools mostly cannot read an Arabic invoice at all.
What I built: I built two of the three layers: the entire AI extraction engine in n8n, and the complete frontend. The frontend is three role-based applications — admin, accountant, and employee — sharing one React 19 and TypeScript codebase, with an Arabic-default bilingual UI, full RTL through logical CSS properties, and a custom i18n layer. It includes a guided upload flow with in-page camera capture, a receipt browser with server-side filtering, Excel export, and feature gating across subscription tiers, all shipped through GitHub Actions quality gates and atomic SSH deploys.
My role: AI/automation engine and complete frontend. Laravel backend by Haitham Zedan (opens in a new tab).

Timed with ZATCA e-invoicing expansion in Saudi Arabia and mandatory JoFotara in Jordan — every paper invoice in the region now has to become clean digital data.

Internal platform2025–2026

DFS Dashboard

A unified CRM and quotation platform that replaced scattered manual processes across every department.

The problem: Customer management and quotation handling lived in separate manual steps with no shared context, so nothing was traceable end to end.
What I built: I owned the two core frontend modules — CRM and Quotation — and connected them into one workflow: creating and updating quotations, tracking statuses, managing customers and leads, and an internal module for projects and tasks with assignment and progress tracking. I worked closely with the backend team on API definitions and payload reviews so UI behaviour matched the business workflow.
My role: Frontend lead

Mobile app · Google Play2025

GoldenTag

A jewelry store management app shipped to the Google Play Store.

The problem: Jewelry retailers track inventory, staff, and sales on paper, in a market where the exchange rate moves daily.
What I built: I coordinated delivery end to end: requirements with the client, scope and acceptance criteria, user stories on Trello, and oversight across frontend, backend, and UI/UX through to publication. Features include inventory, employee management, barcode scanning, USD/SYP exchange rates, sales logging, and TXT export.
My role: Project coordinator

AI web app2026

Business Card Extractor

Photograph a business card, and it lands in your phone's contacts in one tap.

The problem: You collect a stack of business cards at an event and never open it again. It is a small problem that quietly costs sales teams real pipeline.
What I built: A web app that turns a photo of any business card into structured data — name, company, email, phone, website — in seconds, stores it in Google Sheets, and generates a vCard so the contact saves straight to the phone. No custom backend at all: an n8n workflow receives the image, GPT-4o Vision reads it, the data is validated and stored, and it flows back to the app. The one-tap save came from a user suggestion after the first release.
My role: Full build — frontend and automation

AI tooling2026

MCP Search Server

A Model Context Protocol server that lets an AI model query real data instead of only generating text.

The problem: Language models are fluent but disconnected. MCP is the bridge — but most of the discussion about it stays theoretical.
What I built: I built an MCP server with custom tools including a search tool, wired it to real data, and connected it to Claude — so a plain-language question like "what are the details of user 2" returns the correct record directly. The same pattern powers the MCP server that ships with this portfolio.
My role: Full build