React Native Expo Supabase

Solo-built, live on GitHub Releases. Replaces the official FAST-NUCES Flex portal with marks analytics, attendance risk tracking, timetable export, past papers hub, teacher reviews, GPA calculators, and more — built late nights between exams.

highlights

• 📊 Live marks diff — notified when new marks post
• 🎯 Attendance risk + bunk/needed calculators
• 📚 Materials hub — past papers, notes, outlines
• 🧮 SGPA/CGPA approximation the portal won't give you
• 🎨 Custom accent colors, font scaling, concise UI mode

Flutter Supabase Libsodium

A ground-up Material 3 redesign of the open-source Note Safe app. End-to-end encrypted, local-first, with a fully overhauled UI — asymmetric chat bubbles, panorama-aware media viewer, per-group privacy shield, biometric grace periods, and Google Fonts integration.

highlights

• 🎨 Full Material You dynamic color engine
• 🔒 App-Lock Shield + screenshot blocking
• 👁 Per-group blur privacy shield
• 📸 Edge-to-edge media with 10× pinch zoom
• 🔤 Custom typography via Google Fonts

React Node.js Express MSSQL

Full-stack web platform for managing university academics, events, and campus resources. REST backend with a student-first frontend.

Java JavaScript CSS

Web app for university flex management — streamlined UX over the default portal with cleaner data flows and improved accessibility.

C# WinUI

Native Windows companion to the web platform. Offline-capable, faster, and purpose-built for the desktop workflow.

React Native

Android app for managing multiple simultaneous timers. Reliable notifications, clean UI, zero bloat.

Python HuggingFace Flan-T5 mT5

Systematic evaluation of linguistic robustness in small LLMs across 3,600 paraphrase-output pairs. Tests 4 paraphrase types × 2 models × 3 tasks. Key finding: inconsistency is a model-level property, not input-driven.

Python scikit-learn OULAD

Unsupervised clustering + anomaly detection (LOF, Isolation Forest, DBSCAN) on 32,593 students to identify withdrawal risk with no labelled training signal. LOF F1: 0.305 — 1 in 3 flagged students genuinely at risk.

Python OpenCV NumPy

Classical CV pipeline — Harris corners + pyramidal Lucas–Kanade optical flow. User draws a bounding box; tracker propagates it across frames with loss detection and greedy re-detection. No training data required.