Software development · AI agents · Backend systems
Auckland, New Zealand · Master of IT, First Class Honours
I'm an Auckland-based software developer with a Master of Information Technology from the University of Auckland, completed in September 2026 with First Class Honours.
I build Python services, React/Next.js interfaces and AI-agent automation. My work focuses on explicit permissions, clear failure handling and evidence that distinguishes attempted actions from verified results.
During my University of Auckland Industry Internship, I contributed frontend components, REST API integration, prompt/search logic, testing and documentation within a university team.
I'm open to graduate and junior software, frontend/backend and AI/automation roles across New Zealand, including relocation. I hold a three-year post-study work visa; no sponsorship is required during its validity.
Amadeus — 8 merged PRs in Python, TypeScript and C#, verified 24 October 2026.
- MCP approval mapping, environment forwarding, permission-state publication and agent narration evidence.
- Cancellation-safe Pi/MCP cleanup: #226 added three regressions that failed before repair and passed afterward within a related 28-test suite.
- Persistent approval policy and exact-call MCP grants with retained permission boundaries: #234.
- Phase-aware wallpaper rollback and process-identity-bound restart recovery: #228. A separate host-protocol limitation remains documented.
Extension update/rollback proposal #261 adds exact-version startup scheduling and cleanup evidence. Pending merge; no released Host capability is claimed.
Agent GUI Runtime / Agent Review Instant
Released Windows execution runtime for MCP-compatible agents: visual grounding, evidence-bound input, screenshot verification and structured receipts. Execution 0.1.8 adds confirmed updates, task-aware maintenance locks and rollback.
Current-agent mode supports basic clicks without a separate Runtime vision/Decision API. Scoped 0.1.7 evidence includes 20 live 7-Zip steps and a separate installed Calculator workflow. The 0.1.8 development regression group recorded 502 passes and two skips. These records do not establish general unattended reliability; learning is paused.
Python · FastAPI · HTTP/MCP · Computer vision · pytest
Evidence-Driven Trading Desk
A clean-room Java domain core for event-driven trading systems: execution-based accounting, inventory-aware quoting and aggregate risk controls. Its 46 focused JUnit tests run independently of exchange connections. Historical trading-challenge tests and simulations are separate.
Java 17 · Maven · JUnit · Event-driven systems
- IdeaSense AI coursework prototype: implemented most of the frontend in a five-person University of Auckland CS778 team: workspace, staged conversations, summary confirmation and report presentation. This covers the original coursework prototype.
- Networked game server and browser chess client: C#/.NET 8 and JavaScript, with shared state, input validation and concurrency controls.
- Data and diagnostics: Python analysis with pandas/scikit-learn, plus HTTP API and DNS troubleshooting exercises.
More context, case studies and verification boundaries are on my portfolio.
- Languages: Python, TypeScript, JavaScript, Java, C#, SQL
- Frontend: React, Next.js, Tailwind CSS
- Backend and systems: FastAPI, REST APIs, MCP, .NET, TCP sockets, asyncio
- Data and AI: pandas, scikit-learn, OCR, vision-language models, structured LLM outputs
- Verification and delivery: pytest, JUnit, Jest, Playwright, Git, GitHub Actions, Docker