Full-stack engineer — I build and ship production systems, not demos.

I work across the whole stack — from RTL-first React interfaces down to queue-backed NestJS services, Postgres schemas, and the Docker/Caddy setup they run on. Most of what I build touches real money, real users, or real clinical decisions, so I care a lot about state machines, idempotency, and what happens when a third-party API times out.

Alongside that I do applied machine learning: gradient-boosted and ensemble models for screening and classification problems, with the evaluation work that makes them defensible rather than just accurate.

TravelHub (Safariyat) — Arabic-first flight booking platform, live in production

Search → book → pay → e-ticket, end to end. Live flight inventory through Duffel, real card payments through Paymob, server-rendered bilingual e-ticket PDFs, refunds and cancellations, and a role-gated admin console for operations.

The booking lifecycle is an explicit state machine with queue-backed fulfillment and webhook-driven reconciliation, backed by a double-entry ledger — so a dropped webhook or a half-completed payment resolves to a correct state instead of a support ticket.

Next.js 16 React 19 NestJS 11 TypeORM PostgreSQL 16 Redis BullMQ Docker Caddy Azure

Lexora — Gamified dyslexia screening for children, no special hardware

33 tasks across seven cognitive domains, producing a risk signal in under 15 minutes — aimed at the gap between when a child starts struggling with reading and when they actually get a clinical evaluation.

Separate Random Forest models per developmental band (7–8, 9–11, 12–17), with decision thresholds tuned via Youden's J to favour sensitivity, since a missed case costs far more than a false alarm. Only anonymous numerical interaction events leave the browser.

Next.js 15 FastAPI Python 3.12 scikit-learn Vercel

Market Pulse — Financial-analytics landing page, zero UI dependencies

Built deliberately without Tailwind or any component library — vanilla CSS Modules, centralised design tokens, and hand-written keyframe animations. An exercise in owning the styling layer instead of renting it.

Next.js 16 React 19 TypeScript CSS Modules

Dental Model — LightGBM clinical prediction model with a Streamlit interface

A trained gradient-boosting model served through a small Streamlit app, built so a non-technical clinician can actually use it.

Python LightGBM Streamlit

Sentiment analysis on Steam review data, PageRank from scratch, quickselect vs. merge sort benchmarking, feature selection and extraction tooling, distributed-systems and networking coursework, and a run of smaller React and Next.js builds. The older repos are where I was learning; the recent ones are where I was shipping.

Languages — TypeScript, JavaScript, Python, SQL, HTML/CSS

Frontend — Next.js (App Router), React, Tailwind CSS, CSS Modules, RTL/i18n

Backend — NestJS, FastAPI, TypeORM, REST APIs, BullMQ, OAuth 2.0 / JWT

Data — PostgreSQL, Redis, pandas, scikit-learn, LightGBM

Infra — Docker & Compose, Caddy, Azure, Vercel, GitHub Actions