The Problem

The client operated a gym chain relying on spreadsheets, physical logbooks, and manual payment receipts. New member onboarding took 15–20 minutes per person, access control was handled by staff manually checking IDs, and trainers had no dedicated channel to communicate with members.

The Solution

Architected a multi-tenant platform across web (Next.js 15) and mobile (React Native / Expo) backed by a single Supabase + PostgreSQL database with row-level security. Integrated Revenue Monster and Razorpay for payment processing, a QR-code generation system for keycard-free gym entry, and a real-time chat layer for trainer–member messaging.

The Problem

A property development firm needed their website rebuilt, but had no technical team to manage content updates. Every text change required contacting the original developer and waiting days for deployment.

The Solution

Built a decoupled architecture: a statically generated Next.js public site (ISR for near-instant rebuilds) and a private admin CMS protected by JWT auth with rate limiting. All models are database-driven and fully CRUD-capable from the dashboard.

The Problem

Malaysian smallholder papaya farmers lack affordable, real-time access to crop disease diagnostics. Agronomist consultations are expensive and slow — by the time a diagnosis is made, crop damage has spread significantly.

The Solution

Trained a custom EfficientNetB0 model on a curated papaya disease dataset for multi-class image classification. A second stage pipes the classification result into a Hugging Face LLM that generates plain-language treatment recommendations in Bahasa Malaysia and English.

The Problem

The Johor state education department managed dozens of school TV stations (TVPSS) through a fragmented system of emails, Excel sheets, and phone calls. Equipment tracking was unreliable, and student crew management had no audit trail.

The Solution

Designed a layered Java Spring Boot application with Hibernate ORM and a MySQL database. Three distinct role hierarchies (state admin, school admin, teacher/crew) each see tailored dashboards. RESTful API endpoints support scheduling, inventory CRUD, and a bulk-onboarding CSV import for student crews.

The Problem

Job seekers waste hours manually tailoring emails and resumes for each application, with no systematic way to track which roles they've applied to or which documents they sent. Traditional OCR tools require heavy system dependencies, making the setup painful.

The Solution

Built a decoupled FastAPI + React.js platform that accepts job descriptions and resumes through any format — text paste, PDF upload, image upload (OCR via Gemini vision), or a direct URL. Gemini 2.5 Flash generates tailored emails and ATS-optimised resume rewrites. A SQLAlchemy-backed application tracker with Kanban and list views manages the full job application lifecycle.

The Problem

Enterprise operations require programmatically interfacing with WhatsApp messages to synchronize states and logging telemetry to platforms like Notion. However, modern bot mitigation systems employ sophisticated behavioral fingerprinting (analyzing navigator.webdriver flags, WebGL capabilities, mouse movements) that instantly block standard headless automation engines. Additionally, real-time sync introduces a high risk of duplicate records or data gaps during API rate limits and network blips.

The Solution

Developed a Python-based headless browser automation engine powered by Playwright and Selenium, featuring a modular pipeline architecture with isolated boundaries. The engine includes a stealth injection layer that dynamically overrides client-side JavaScript automation properties and simulates human-like timing variations to evade detection. To handle network blips, the pipeline uses an isolated filter layer and a localized exclusion system (excluded_log.txt) for robust state reconciliation without duplication.

The Problem

Traditional educational facilities rely on outdated, manual grading spreadsheets and disjointed, error-prone academic record keeping. This fragmentation slows down advisor intervention loops, makes real-time student performance comparison extremely difficult, and leaves student grade-remarking workflows opaque and unmonitored.

The Solution

Engineered StudyFlow, a multi-tenant academic ERP designed with a decoupled client-server model. It implements dynamic Vue 3 dashboards tailored for administrators, lecturers, academic advisors, and students. The backend utilizes a Native OOP PHP REST API with Custom JWT authentication middleware, orchestrating grading pipelines, peer matrices, and re-marking loops inside a containerized Docker-Compose stack.

The Problem

Traditional multi-tenant file indexes rely on heavy application servers and complex routing stacks that create massive memory overhead. Enforcing granular dynamic permissions across institutional semesters, categories, and direct links via traditional ORM queries introduces severe database bottlenecks and security privilege escalations.

The Solution

Engineered StudySync, utilizing an un-nested, low-overhead custom Perl 5 TCP daemon that parses socket headers manually to serve stateless API routes. Built a highly normalized MySQL permission schema mapping semester, category, and direct link matrices, and paired it with a decoupled, reactive Vue.js front end. This shifts access-control verification to efficient single-step database joins, keeping idle server memory footprint minimal.