Things I've built.

Analysed 6,600 student records to identify key factors behind exam performance. Developed and compared Linear Regression, Decision Tree, and Neural Network models.

ML pipeline to predict hourly ridership for Malaysia's Komuter train system using open data from data.gov.my. Trained and compared five models including Random Forest, XGBoost, and AutoML, with hyperparameter tuning for final selection.

Credit risk assessment system combining structured financial data with unstructured borrower text. Uses XGBoost for numerical features and an LLM+RAG pipeline to extract behavioural signals from loan essays and bank statements.

Predicts individual medical insurance charges from personal attributes, built to understand model behaviour rather than chase scores. Compared seven regression models and engineered interaction features around the dataset's three cost bands: linear models plateau at R² 0.87 while Random Forest cuts MAE to around $1,960.

Property and travel listings web app with full CRUD functionality and location-based search filtering. Schema validation with Mongoose and Joi.

Redesigned heritage organisation website as a full-stack app with an AI chatbot (Meta-Llama 3.2) accessible across all pages. Backend built with Spring Boot and Java 21, containerised with Docker.

Real-time messaging app with one-to-one and group chats, live presence, typing indicators, read receipts, and file sharing. Auth is unified across HTTP and WebSockets using signed-cookie JWTs; messages load with cursor-based pagination.

Full-stack AI chat application powered by the OpenAI API. Features persistent conversation threads, file attachments, an incognito chat mode, model switching at runtime, and markdown-rendered responses. Containerised with Docker.

Finance management system with debit/credit transactions, loan management with interest tracking, and automated monthly savings allocation. Secure auth with BCrypt and persistent data via CSV file I/O.