Sonny
Hiu Long
Kam

Shipped production features used daily by property valuers and support staff, building end-to-end flows across a Node.js/Express backend, PostgreSQL/Sequelize, and a Vue.js/Vuetify frontend.

Reduced manual spreadsheet work and improved quote and approval tracking, delivering portfolio valuation workflows with clear business rules, data models, and UI journeys for jobs, contacts, and portfolios using PostgreSQL + Sequelize and Vue/Vuetify.

Hardened access to high-value valuation data, implementing secure auth with email/password, Google + Microsoft SSO, TOTP-based 2FA, session handling, and reCAPTCHA.

Improved data consistency and API reliability, building REST endpoints for valuations, portfolios, messages, and documents with strong validation and error handling in Node.js/Express.

Improved delivery speed and reduced rework, collaborating with founders, valuers, and engineers to clarify requirements, break work into Jira tickets, review pull requests, and document features.

Reduced regressions in core valuation and auth flows, using Git, PR reviews, and a CI pipeline with automated checks to keep releases stable.

Redesigned and rebuilt the company website from an outdated layout to a modern, responsive UI.

Handled domain setup and routing, including DNS updates and redirects to keep existing links working.

Improved SEO by fixing metadata, page structure, and indexing issues, then tracked results in Google Search Console.

Modernised the website with a refreshed design system, cleaner navigation, and mobile-first pages.

Migrated domain and configured routing rules (redirects, canonical URLs) to prevent duplicate pages and broken paths.

Boosted search visibility with on-page SEO updates and Google Search Console monitoring, including sitemap submission and crawl fixes.

Bachelor of Engineering (Honours) in Software Engineering.

GPA: 3.2 / WAM: 75.

Final Year Project: AI Human–Object Interaction Detection in Video Data | Project Lead / Developer • Built an AI system that improved human–object interaction recognition in video data, combining pose estimation with object detection to model motion and interactions. • Converted raw videos into structured training data for learning interactions, extracting 17-joint skeleton (x, y) coordinates per frame and pairing them with detected object coordinates using Python. • Improved interaction detection accuracy by +8% on ST-GCN and +2% on ST-GCN++, redesigning the pipeline and feature fusion approach for stronger spatial-temporal signals using deep learning in Python. • Increased temporal feature quality for complex interactions, testing UniFormerV2 and integrating its temporal modelling with ST-GCN++ to strengthen representation learning. • Protected privacy during inference and demos, applying identity blurring so no personally identifiable details appeared in processed outputs.