Cybersecurity engineer and AI safety researcher based in Auckland, New Zealand. Two-time 3rd-place winner of the Palisade Research 2025 AI Misalignment Bounty — two separate prizes from a single elicitation scenario that produced fundamentally different fabrication patterns from o3 and GPT-5.
Six years across APT threat hunting, SIEM detection engineering, and quantitative analysis. Currently building autonomous incident-response agents for the SANS FIND EVIL! competition — applying frontier-AI evaluation methods to defensive security operations.
MuJoCo-simulated Unitree G1 humanoid on Inspect AI (UK AISI), testing whether AI models deceive operators when controlling a physical robot under pressure. Across 4 frontier models and 30 reproducible runs, found safety and honesty are independent failure dimensions — Gemini Robotics ER 1.5 showed L3 strategic concealment in 29% of runs, while GPT-5 stayed fully honest (5.0/5) despite low safety (1.9/5). Telemetry-corruption variant probes concealment of safety violations. Built at the Gemini 3 Hackathon.
Technologies: Inspect AI, MuJoCo, Next.js, Python
Full-stack auction marketplace with AI-powered features, built with a team of 3 developers and 2 UX designers.
- Real-time bidding, watchlists, product comparison
- AI-powered Q&A system (Gemini/Ollama)
- RESTful API with MongoDB
- 56+ passing tests using TDD methodology
Technologies: React 19, Node.js, Express.js, MongoDB, Google Gemini API, Jest
Conversational AI insurance consultant with multi-turn dialogue and business rule enforcement.
Technologies: React 19, Node.js, Express.js, Google Gemini API, Vite
| Cybersecurity | Frontend | Backend | Cloud & DevOps | AI/ML |
|---|---|---|---|---|
| Splunk SPL | React | Node.js | GCP | OpenAI API |
| SIEM Detection | TypeScript | Express.js | Azure | Anthropic API |
| APT Hunting | JavaScript | FastAPI | AWS | Inspect AI |
| Threat Intel | HTML5/CSS3 | Python | Docker | Gemini API |
| Blue Team Ops | Tailwind | REST APIs | Git / CI-CD | Prompt Engineering |
Databases: MongoDB, Mongoose, SQL Server, PostgreSQL, Supabase, Firestore
Testing: Jest, Supertest, Playwright, Vitest, axe-core, TDD
Infrastructure: Vercel, Resend, Upstash, Next.js App Router
- Director & Technical Lead @ IPU Collective — Māori-led studio delivering for Crown and council clients (Auckland Transport, KiwiRail, Watercare, Oranga Tamariki)
- Full Stack Engineer (Contract) @ Takitoru Developments Ltd — Architecting full-stack mobile + web platform end-to-end with test-first delivery
- Independent AI Safety Researcher @ Approxiom Research — Frontier-model vulnerability discovery and alignment evaluation
- Cybersecurity Engineer @ Workday — APT threat hunting, custom Splunk SPL detection rules, Python automation for threat intel, global SIEM efficacy analysis tool
- Pasifika Tech Tutor / Speaker @ Pasifika Tech Education Charity — Workshops helping Pacific students develop practical tech skills
- Data Analyst @ Halter Limited — Agricultural IoT data analysis, ML-based products
- Actuarial Analyst @ Ministry of Social Development — Long-term welfare outcomes modelling
- AI-Powered Advanced Full Stack Developer (L5) — Mission Ready HQ, January 2026
- Full Stack Developer Certificate (L4) — Mission Ready HQ, August 2025
- BSc (Honours) Statistics — University of Auckland
- BSc Double Major, Mathematics & Statistics — University of Auckland
- Palo Alto Networks Cybersecurity Professional Certificate — Coursera
- Blue Team Level 1 (BTL1) — Security Blue Team
- AWS Cloud Quest: Cloud Practitioner — Amazon Web Services
Two-time 3rd-place winner — Palisade Research 2025 AI Misalignment Bounty
- Published: arXiv:2510.19738 (9 of 295 submissions awarded)
- Winning submissions on Hugging Face:
- Demonstrated that a single misalignment elicitation produced fundamentally different fabrication patterns from o3 and GPT-5 — two prizes from one scenario
