This page brings together links to personal and AI-inspired ideas that I have been exploring through both my research and personal interests. It provides a space to share ongoing projects, experiments, and creative work that connect my academic interests with the technologies and ideas I enjoy pursuing in my own time.
Agentic Video Games of My Childhood
Kuga Invaders — An original Space Invaders-style browser game built with JavaScript, D3.js, and SVG. It combines testable game-state simulation with D3-powered rendering, responsive keyboard and mobile controls, anime-inspired enemies, an Overdrive scoring system, synthesized audio, progressive difficulty, offline PWA support, and automated tests.
CityUber — An elevator-inspired public-transport simulation set on a stylized Port Moresby map. The player manually dispatches two CityLift vehicles while competing against an adaptive rival AI that responds to passenger demand, waiting times, traffic, and score. It features editable stop queues, passenger simulation, traffic-aware routing, accessibility and energy scoring, responsive SVG animation, and deterministic tests.
Kuga Man — An original, asset-free maze-chasing game built with vanilla HTML, CSS, and JavaScript. Players collect data dots and power cores while avoiding four Glitch Guards with distinct targeting personalities. It includes responsive Canvas graphics, keyboard and touch controls, power mode, score multipliers, persistent high scores, sound, and no external dependencies.
Tower Agents — An experimental, mobile-first programming game about operating a living high-rise with AI coding agents. Players describe goals such as reducing waiting time, improving accessibility, or saving energy; specialist agents then propose sandboxed strategy changes, test them through deterministic simulations, explain metric trade-offs, and support diff review, approval, rejection, and rollback.
Personal AI Tools
- PDF Sadoku — AI-assisted PDF Reviewer — A calm, local-first workspace for reviewing PDFs and placing constructive notes directly on manuscripts. It supports precise annotations, separate author-facing and private comments, AI-assisted page review and comment refinement through Pi, annotation filtering, and export to Markdown, JSON, or a flattened annotated PDF while keeping documents and credentials within a privacy-conscious local boundary.
