// currently building things that should exist

Founder
Engineer
Maker

I turn messy ideas into understandable systems.

I like boring foundations, explicit boundaries, observability, deterministic behaviour where possible, and AI only where it earns its place.

Technical lead · senior full-stack engineer · 20+ years · Melbourne, Australia

Engineer

Or just ask this page

01 / things I'm building

Selected experiments

Click a project ↓

quixo / the question

Can software teach the way a great tutor thinks?

A good tutor does not generate a plausible answer — they know where the student actually is, and they can prove it. So the maths stays deterministic and checkable, the conversation carries the teaching, and every session leaves evidence of what was learned.

Why this one

The problem was personal. I have two kids in Years 7 and 10, and when maths got hard we tried everything on the shelf — YouTube, general-purpose chatbots, learning platforms, tutors. Each helped in isolation. None of them joined the journey up: work out where the student is actually stuck, explain it in the way that lands, set the right practice, then show whether the misconception has genuinely gone.

I started by building interactive games that made hard concepts visible — my old game-building habit, pointed at homework. The more I built, the clearer the bigger gap became: students did not need another pile of lessons and games, they needed a system that could diagnose a misconception, adapt the practice and track mastery over time.

Teachers said the same thing from the other side of the desk. They want to personalise, but they have limited time and a room full of different needs — and most platforms will tell you an answer was wrong without telling you why, or what to do next.

Direction

Curriculum-aligned explanations, modelled as data — progress is a position, not a score.
Deterministic step verification — the marking is reproducible, not sampled.
Misconception detection, so the next question is the right question.
Interactive games that make a hard concept visible rather than described.
Mastery tracked over time, and legible to the student, the teacher and the parent.

Built with

TypeScript
React
Next.js
Supabase
Postgres + RLS
LLM adapters

status: in build

02 / operating system

How I like to build

01

Start with the simplest useful thing.

Architecture should earn its complexity, not assume it.

02

Make it observable.

If you cannot see why it behaved that way, you cannot improve it.

03

Give AI boundaries.

Loops, routing, evaluators and human gates beat mystical diagrams.

04

Build the thing.

Software, hardware, product — prototypes answer arguments quickly.

LearnDeck / field notes
Agentic architecture without the mythology.
Most useful agents are loops plus tools, routing, evaluation, memory and carefully chosen human gates. The hard part is knowing when each belongs.

03 / writing

Notes from the build

all of LearnDeck ↗

04 / lab

Half-finished on purpose

This page — The site is the demo — a deterministic intent router answers questions about the work, with an adapter left open for a real model.
Agent playground — A visual agent architecture sandbox where every loop, tool call, evaluator and human gate can be switched on and off live.
Vision on moving things — Small classifiers running on Raspberry Pi hardware, where latency and light are not hypothetical.

tiny terminal

Same knowledge base as the chat, addressed by command.

Try:

$ ready_to_build

Type a command — or just ask a question in plain English.

idea machine

What should we build next?

A visual agent architecture playground where every loop, tool call, evaluator and human gate can be switched on and off live.

05 / about

Twenty years of owning the thing

I am a hands-on technical lead with 20+ years of shipping software across healthcare, fintech and consumer products — Oracle Health, NVIDIA, NAB, Block Ledger and now Wrkr. The through-line is ownership: architecture, delivery, the team, and the pager.

I am AI-native in a specific way. I design production LLM systems with a deterministic core and AI at the edges — tiered model routing, structured outputs, prompt caching, retrieval where it earns its keep — and I build with agentic development workflows daily.

Outside the job I am building Quixo, writing LearnDeck, and making physical things in the workshop. Same loop, different materials.

Senior Full Stack Engineer / Front-end Lead · Wrkr
Dec 2021 — present · Melbourne
- Lead front-end architecture for the customer onboarding SaaS platform, from Figma wireframes to production.
- Architected it on micro-frontends so a single codebase is themed and deployed per customer — white-labelling without a fork.
- Own the shared design system (Headless UI, Tailwind, theming) that keeps every white-labelled deployment consistent.
- Deliver the API layer behind it — microservices in Java, NestJS and Node — and own AWS across dev, staging, sandbox and production.
Senior Analyst, Engineer · NAB
Oct 2019 — Oct 2021 · Melbourne
- Owned solution design for NAB One, the bank-wide intranet platform serving staff across the enterprise.
- Quality gatekeeper for the codebase — review, coverage standards, PR approval — plus enterprise-wide reusable patterns.
- Drove spikes, POCs and root cause analysis; established TDD and API-first practice.
Lead Engineer · Block Ledger
Jan 2019 — Oct 2019 · Melbourne
- Led the Luca+ accounting and auditing platform end to end — architecture, build and delivery.
- Containerised microservices (Node, Express, MongoDB, Docker, Kubernetes) with documented APIs; led the migration to GCP and AWS.
- Built ML-driven risk scoring that surfaced financial risk from client invoice data.
Full-Stack Architect · NVIDIA
Feb 2016 — Dec 2018 · US, India
- Shipped an ML recommendation engine that lifted engagement 150% across the GeForce Experience user base.
- Built ML-based bug auto-triage on the ELK stack (+30% team productivity) and eliminated the top 10% of crash-causing issues.
- Prototyped and productionised across the life cycle — deep-learning personalisation, React Native hybrids, Node native add-ons.
Lead Architect · Oracle Health (Cerner)
Oct 2004 — Feb 2016 · US, India
- Led a team delivering continuous deployment of a SaaS healthcare platform with 1M+ users, on 30+ Unix servers at 99.9% uptime.
- Shipped clinical-safety algorithms with direct patient impact — medication reconciliation cut manual errors 80%.
- Modernised the platform: legacy-to-cloud migration, SVN-to-Git, CI adoption across a global team.

Stack

Front-end: React · Next.js · TypeScript · Micro-frontends · Design systems · Tailwind CSS · Headless UI
Back-end: Node.js · NestJS · Java · Python · Microservices · REST API design · Postgres · MongoDB · Supabase
Cloud & platform: AWS (EKS, EC2, RDS, S3, IAM) · GCP · Docker · Kubernetes · CI/CD · Linux
AI engineering: Tiered model routing · Structured outputs · Prompt caching · RAG where it earns it · Deterministic core / AI periphery · Agentic dev workflows · Recommenders & triage · Computer vision
Physical: Raspberry Pi · 3D printing · Drones · Robotics · Laser cutting · Electronics

Education

2018: PG Diploma, Machine Learning & Artificial Intelligence — International Institute of Information Technology
2004: Bachelor of Engineering, Information Science — Visvesvaraya Technological University

Ways in