Handpicked.

The frontend — React, Vite, TypeScript, and Tailwind — carries a clean luxury aviation aesthetic: dark gradients, gold accents, and smooth motion that matches the world of private travel. A lightweight Node.js + Express backend connects to RapidAPI’s Skyscanner endpoints and delivers normalized, real-time flight data. Vercel hosts the frontend, Railway powers the backend, and together they form a seamless, production-ready pipeline.

RapidAPI’s rate limits demanded careful fallback logic and smart response shaping to keep the experience smooth. Airport inputs had to be validated and normalized, because users type city names instead of IATA codes. The luxury UI had to stay consistent across laptop, tablet, and mobile — every gradient, shadow, and motion intentional — while CORS, environment variables, and production API stability all needed solving.

One-way, roundtrip, and multi-city searches return results that feel like products: aircraft category, seat count, range, operator, availability window, and indicative pricing. A compact mobile card layout keeps browsing effortless on small screens. Fast, elegant, and ready to evolve — a clear proof of concept for investors, operators, and premium travelers who want clarity in a space that rarely offers it.

Two deployments of one identity. The Exhibition Version is a visually polished, frontend-only build for academic presentations and conferences, running on mock data to demonstrate the full flow without exposing any real student information. The Production Version is the real-world implementation: Firebase Authentication, Firestore real-time data, protected dashboards, and role-based access for students, counselors, and parents.

One system had to serve two completely different deployment contexts while keeping identical UI and user flow. The exhibition build had to look fully functional using mock data, and the production build required real authentication, encrypted data storage, and role-based access controls — without the two ever drifting apart.

A dual-version architecture sharing the same React codebase, switched by environment-based feature flags. Exhibition mode runs realistic mock data with believable loading states for conference demos; production mode integrates Firebase Authentication, Firestore, and role-based routing. Structured psychological data models were designed in collaboration with the PhD researcher to capture every required metric and evaluation criterion.

Kevo transforms raw user inputs into a complete professional profile and delivers personalized, data-driven career recommendations. It helps users analyze their personality and cognitive style, discover career paths that match their strengths, and understand why certain roles fit them better — making AI guidance accessible and meaningful for students, professionals, and career changers.

Building an AI reasoning engine that felt intelligent and personalized — not like a generic quiz — required careful prompt engineering and structured output parsing. Input validation, rate limiting, and a secure contact system were also required to make the platform production-ready.

React + Vite for fast performance, Framer Motion for cinematic animation, and TailwindCSS for clean styling. Structured prompts analyze user inputs and generate detailed career recommendations with explanations, while a secure Nodemailer contact system — protected by rate limiting and input validation — lets companies reach out directly. Intelligence, turned into action.

A modern fitness landing page designed specifically for the Lebanese / Baalbek market — bold, energetic, and culturally aligned with the identity of local gym owners. The goal was a concept website that demonstrates strong branding, clean UI/UX, smooth animations, and a theme inspired by local fitness culture.

The design had to feel globally modern and locally relevant at once — meeting contemporary web standards for animation, responsiveness, and branding, while still appealing to gym owners who may never have had a digital presence before.

A fully responsive React website: hero with strong branding, class categories, trainers, pricing plans, contact, and smooth scroll animations. Bold typography and energetic gradients carry gym culture while keeping professional polish — commercial-ready branding and UI/UX direction for a specific audience.

A fully structured command-line habit tracking system built in Python. The focus is clean architecture, functional programming, and reproducible behavior — a strong demonstration of backend logic and software design principles, managing habits, tracking progress, and analyzing streaks without any UI framework.

Most habit-tracking tools are visual and heavy. This one had to be lightweight and architecture-driven — every feature implemented through disciplined terminal I/O with robust data handling underneath.

Clean architecture separating concerns into distinct layers, functional programming for predictable, testable behavior, habit creation with configurable periods, check-in tracking with streak calculation, progress analytics, and persistent SQLite storage. Real logic, not just interfaces.

A full-stack predictive maintenance platform that monitors industrial machines, detects anomalies, and predicts failures from real-time sensor data. Mock sensor data — 2000+ readings across temperature, vibration, pressure, humidity, operating hours, and load — simulates realistic industrial IoT and can be replaced with live data from any company.

Simulating realistic industrial IoT data while classifying machine status into ok, warning, and critical thresholds; orchestrating Gemini, DeepSeek, and Groq with careful fallback handling; and triggering alerts the moment failure probability passes 0.7 — all behind a security-hardened backend with Helmet, CORS, rate limiting, and input validation.

A RandomForestClassifier with 200 estimators and IsolationForest anomaly detection, served through a Node.js + Express API with cached lookups. A React + Tailwind + Framer Motion frontend with lazy loading and code splitting, Supabase storage, SendGrid alerting, and a containerized AI service running Gunicorn with 4 Uvicorn workers — deployed on Railway and Vercel.

A complete rebuild of a 3D-driven hero section designed to deliver a cinematic first impression through precise alignment, responsive behavior, and a clean visual hierarchy — a centered, single-line heading paired with a floating 3D portrait that remains fully visible, never clipped, duplicated, or misaligned across devices.

Preventing text wrapping on large screens, maintaining consistent spacing, and making the 3D portrait adapt smoothly without breaking the theme or color system at every viewport size.

Precise CSS and component architecture. Every breakpoint was tested and tuned, the 3D integration carefully aligned, and the color system enforced consistently — a pixel-perfect, production-ready interface that preserves the cinematic original while demonstrating strong debugging, design-system enforcement, and responsive skills.