A review platform for assessment teams, covering submission review, exam version comparison, report navigation, and rule-based suggestion generation.

• Built and integrated backend REST API endpoints powering the core review workflows — submission review, exam version comparison, report navigation, and rule-based suggestions.
• Refactored the frontend from a legacy Vue implementation to React 19 + Vite, restructuring pages into feature-based modules and reusable UI components.
• Owned technical documentation, release notes, and test evidence across the team's GitHub workflow.

• React 19
• Vite
• TypeScript
• Node.js
• REST API
• GitHub

Case study

A map-based explorer for the City of Melbourne Urban Forest dataset, supporting search across tree locations, species groups, maturity status, and nearby landmarks.

• Automated a Python pipeline that cleans, transforms, and converts 82,064 spatial records into map-ready marker data, removing the manual preparation step entirely.
• Designed custom SVG map markers per tree category to keep a dense, large-scale dataset legible.
• Implemented spatial filtering and querying by distance, species, and maturity status for location-based analysis.
• Added user-generated markers for significant locations to support interactive trip planning.

• Python
• Pandas
• WordPress
• WP Go Maps
• SVG
• Geospatial

Dataset

An end-to-end preprocessing pipeline for predicting mortality in acute kidney injury patients, built on the MIMIC-IV critical care database.

• Defined an AKI cohort in BigQuery, joining ICU stay details, KDIGO AKI stages, severity scores, labs, vital signs, blood gas values, and comorbidities.
• Built a reproducible Python pipeline covering cohort cleaning, missing-value imputation, outlier handling, log transformation, normalization, and categorical encoding.
• Analyzed 61,742 ICU stays with exploratory data analysis, correlation analysis, and principal component analysis.

• BigQuery
• SQL
• Python
• Pandas
• NumPy
• PCA

A PyTorch pipeline that estimates the calorie content of a meal from paired colour and depth imagery.

• Built a preprocessing pipeline for 3,086 RGB-D food image records, covering loading, augmentation, label normalization, and model preparation.
• Automated an 80/20 train–validation split so experiments stay reproducible across runs.
• Evaluated RGB against RGB-D models; the final fusion model improved validation MAE from 69.7 kcal to 59.6 kcal.

• PyTorch
• Python
• Computer Vision
• RGB-D
• Data Augmentation