Nik Zulfaqar

Building scalable systems and GenAI-powered solutions

I am an with several years of experience building and optimizing scalable applications, including work on developing GenAI-powered solutions. Skilled across React, Vue, GraphQL, REST APIs, Flask, Node.js, and Express.js, I consistently deliver clean, efficient, and reliable systems end-to-end.

View Projects Resume

Experiences

A chronological record of my career and key contributions.

Associate Software Engineer

Mesiniaga Berhad · Contract

MAR 2024 — PRESENT

Built reusable AI harnesses and context-engineering workflows adopted across engineering departments, structuring codebase context, documentation, and engineering decisions while cutting development effort by 90% against ad-hoc AI prompting.
Built and deployed a machine learning–based workforce absence-prediction platform single-handedly for shift supervisors and line managers, achieving 90% accuracy with LightGBM and 49+ engineered features, cutting backfill planning from a week to minutes.
Scaled the company's internal GenAI platform to 500+ employees across 10 departments, establishing enterprise AI policy governance, executing a zero-error MongoDB-to-PostgreSQL migration, and adding LightRAG, Redis caching, and usage analytics.
Eliminated a 4-business-day vendor dependency blocking finance and presales by reverse-engineering an undocumented carrier billing rule and replicating its behavior in a FastAPI-based postpaid billing system.
Reduced project discovery for proposal and presales writers from 30 minutes of manual folder searching to seconds with natural-language hybrid search, combining vector and keyword retrieval with LLM-based relevance filtering and anti-hallucination safeguards.
Owned deployment and maintenance of 4 production applications on GCP and automated image builds and GHCR pushes to save engineers 1–2 hours per deployment cycle.

FastAPIPythonLangChainDSPyLLM IntegrationRAGMCPAgent EvalsAgentic WorkflowsPrompt EngineeringContext Engineeringn8nMachine LearningLightGBMXGBoostCatBoostFeature EngineeringFlaskWebSocketsETL PipelinesVue.jsReact.jsNext.jsTailwind CSSHTMLPostgreSQLMongoDBRedispgvectorEmbeddingsSemantic SearchVector SearchHybrid SearchDockerGoogle Cloud PlatformDeploymentGitGitHubESP32

Software Engineer Trainee

Mesiniaga Berhad · Internship

AUG 2023 — FEB 2024

Digitized the hospital's maternal health record (pink book) and replaced Excel-based HR attendance with NFC card check-in, building 13 backend modules ahead of the planned schedule and securing data with 66 GraphQL Shield permission rules, with zero security incidents.
Migrated backend APIs from REST to GraphQL, eliminating over-fetching through client-requested field selection and reducing response payloads by 40%.
Built a proof-of-concept Automatic Number Plate Recognition (ANPR) system to replace the guardhouse's handwritten vehicle logbook, achieving 100% plate recognition across 50 simulated vehicles using YOLOv8 on live video.

Node.jsJavaScriptGraphQLMongoDBComputer VisionGit

Skills

A curated stack of tools and technologies I use to build robust digital solutions.

nikzr@dev: ~/skills

nikzr@dev:~/skills$

ai-genai10 packages installed

Designing and shipping Generative AI solutions with modern LLM frameworks, retrieval, and agentic workflows.

LLM Integration
LangChain
DSPy
RAG
MCP
Agent Evals
Agentic Workflows
Prompt Engineering
Context Engineering
n8n

Certificates

Professional courses, vendor training, and achievements across cloud, machine learning, and AI engineering, alongside recognition earned on the job.

nikzr@dev: ~/certificates

nikzr@dev:~/certificates$

mesiniaga1 certificate found

Mar 2026
MAHA Gold Award 2025 (Mesiniaga Award for High Achievers)

Projects

A curated collection of recent work focusing on scalable architecture, interactive experiences, and clean code.

Kimiaroom

An interactive chemistry lab for Malaysian SPM students. Learners search and filter a periodic table of all 118 elements, add substances to a reaction chamber, and run an experiment with a 2D visual simulation, then see the products, balanced equation, reaction type, safety notes, and a cited source. Outcomes come only from a curated, reference-backed reaction catalogue covering neutralisation, gas production, displacement, precipitation, and catalysed decomposition, so the app never guesses chemistry; unsupported combinations say so plainly. React, TypeScript, Vite, and Tailwind CSS on the frontend; FastAPI on a Python backend organised into feature modules, tested with pytest and managed with uv.

ReactTypeScriptTailwind CSSPythonFastAPI

FurtherXpert

An AI-powered course recommendation system that matches SPM graduates with university programmes based on their grades, strengths, interests, and preferences, checking each match against UPU entry requirements. It replaces guesswork with informed study choices: students get a personalised course shortlist, an AI-written summary of their results, and a clear reason for each suggestion. SPM grades and a 20-question interest test become two spider charts showing what they're likely to excel in and likely to love, and every public-university (UA) diploma is labelled Eligible, Conditional, or Not Eligible with reasons. React (Vite), Tailwind CSS, Recharts, and React Router on the frontend; FastAPI, SQLAlchemy, Pydantic, Alembic, and PostgreSQL on the backend, with JWT auth and argon2 password hashing. The scoring and eligibility rules are deterministic Python covered by pytest golden tests; OpenAI GPT-5 only writes the plain-words summary. Runs on Docker Compose.

ReactTailwind CSSPythonFastAPISQLAlchemyPostgreSQLDocker

Jobzone

A web-based tool that flags fraudulent job postings for Malaysian job seekers, helping them avoid scams before applying. An XGBoost model classifies each posting as genuine or fraudulent, automating the initial screening job seekers would otherwise do by hand, and reached 80% accuracy on a test dataset. Nuxt.js, Tailwind CSS, and daisyUI on the frontend; Flask and MongoDB on the backend. Built for DevHack 2023.

Nuxt.jsPythonFlaskXGBoost

Screenshot of the live Infohunt eQuiz showing a multiple-choice quiz question

MSU iReX 14.0 Infohunt eQuiz

An online quiz built as a checkpoint for MSU iReX 14.0 InfoHunt Challenge participants, testing understanding of each team's chosen final year project across 10 questions. Hosted on GitHub Pages with QR code entry.

JavaScriptHTMLCSS3

EyeBuddy

A dry-eye syndrome detection mobile app that implements the CNN-based Break-Up Time (CNN-BUT) algorithm in Python and OpenCV behind a Flutter frontend. Translating the computer-vision algorithm into a mobile app enables portable dry-eye screening, reducing reliance on bulky clinical equipment for preliminary assessments in rural areas. Built for the Apps Innovation Challenge (AIC) 2022, university category.

FlutterPythonOpenCVFirebase