
Featured project
WhatsApp AI Receptionist
A multi-tenant SaaS that puts an AI receptionist on a wedding vendor's own WhatsApp number — grounded answers, tentative holds, and one dashboard for bookings, deposits and finances.
computer science
+ real world impact
Hi, I'm
Fullstack Developer
I'm a Computer Science graduate from the University of Sheffield with hands-on experience building full-stack web applications, and a growing interest in using machine learning to solve real-world problems — from business automation to understanding air pollution sources.
Hi! I am Nur Izfarwiza, but people call me Efa. I'm a fresh Computer Science graduate from the University of Sheffield, originally from Malaysia.
This space showcases the personal and academic projects I've built — from full-stack web development and IoT wearables to machine learning research. I love building things that live at the intersection of creative problem-solving and solid engineering.
My dissertation explored practical ML at scale with the Advection Gaussian Process framework, earning a Distinction (75). I care deeply about writing code that is not just functional but clean, maintainable, and thoughtfully architected.
Products and client work I designed, built and shipped.

Featured project
A multi-tenant SaaS that puts an AI receptionist on a wedding vendor's own WhatsApp number — grounded answers, tentative holds, and one dashboard for bookings, deposits and finances.

A mobile-first meal planner and fitness tracker that turns the meals you already cook into a weekly plan, a ready-to-shop grocery list, and a personalised calorie target — with Gemini filling in ingredients, nutrition estimates and activity calories along the way.

An academic website for Prof. Dr. Mohd Talib Latif that keeps his publication record, citation metrics, media archive and research team current. Built as commissioned client work, with live academic data and a password-gated admin for the records he maintains.


A smartwatch that turns daily habits into something you can watch grow. Tracks steps, nudges you when you go inactive, and grows a virtual plant as you log water.
Two employed roles, both spent on production systems — modernising a platform that could not be changed safely, and building AI automation that had to answer customers correctly.
eAuto Sdn Bhd — eAuto Platform (Malaysian vehicle transfer / JPJ-integrated)
A platform for Malaysian vehicle ownership transfers, integrated with JPJ. The core is a legacy Java 6 / Spring 2.5 / Hibernate 3 / Tomcat 6 monolith, with satellite Spring Boot services for search, reporting, cron, notifications, onboarding, e-invoicing, SSM and asset storage. I worked here as a Software Engineer on both the modernisation and the customer-facing product.
The oldest features were locked inside a 20-year-old Java codebase, so every change carried risk. I architected and deployed `backoffice`, a Kotlin / Spring Boot 3.5 satellite host that extracts features one at a time, and enforced the module boundaries with ArchUnit assertions in CI so they could not quietly erode.
Staff needed a single sign-in, and a sign-out in one app had to take effect everywhere. I built a JWT hand-off with single-use signed tokens, JTI replay guards and an open-redirect whitelist, plus a back-channel SessionRegistry trigger that revokes the other sessions at the same moment.
The nightly reconciliation batch could match a payment against the wrong date, which produced mismatches and deadlocks in money-critical flows. I reordered the engine so match logic resolves before any write happens, and folded the state flips into the existing transaction boundaries rather than opening new ones.
Customers needed to move a car interstate, and the platform had to carry payments and regulatory invoicing end to end. I led a multi-round build across 4 repositories: a three-step customer portal wizard, FPX/RHB payment gateway integration, and RabbitMQ event pipelines for LHDN e-invoicing write-backs.
Drivers authenticate and hand over vehicles at pickup and drop-off, and files need to sit in local storage during development but in S3 in production. I designed dual-token authentication with independent pickup and drop-off handshake lifecycles, and built an S3 SDK storage router that switches between MinIO and AWS S3 at runtime.
Forgotten-password recovery was a weak point in the old monolith. I built a fail-closed recovery flow with MD5-hashed OTPs and 24-hour rate limiting, and wired a session-listener trigger so a reset invalidates every active session rather than only changing the credential.
The D Group
Commercial e-commerce platform work, spanning the retailer-facing catalogue, the support tooling, and the automation layer behind both.
The catalogue had grown past 10,000 rows and the client-side fetches could not keep up. I moved the data access onto Prisma with server-side pagination, so a list view loads a bounded page instead of the whole table.
The same product was listed under several names, which made every downstream report disagree. I built a bulk SKU management flow using Pinecone vector search to match product aliases and flag duplicates, so the catalogue converges on one consistent record.
Customer Service modules were only meant to be reachable by the right people, and the rule varied by context. I built a JWT-based authentication gateway with dynamic routing that resolves access from the request context — including URL parameters — and enforces conditional access centrally, rather than leaving it to each screen.
Retailer questions were answered by hand and retailer verification was tracked manually. I architected a RAG-driven AI agent in n8n, wired to the internal Drive API, that answers end-user questions over WhatsApp, and automated the retailer verification flow so the results feed straight into Jira sprint tracking.
2022 — 2025
Focus areas: full-stack systems, data-driven computing, and machine learning
Have a project idea, collaboration opportunity, or just want to say hello? Feel free to reach out — I'm always open to interesting conversations.