AI-native solutions engineer · Canberra, Australia
I take a business from first conversation to live product on my own. AI coding agents do the build; I own the problem, the decisions, the testing and the outcome.
Case study
A biotech company sells health products through dealers and directly to consumers on WeChat. It ran on disconnected tools, and every marketing change needed outside help. I was the only person on it, from first conversation to daily operations, with AI coding agents writing the code.
How I build
Everything in the case study was built by AI coding agents working under my direction: Cursor from the first commit, Claude Code since October 2026. I decide what gets built and how it should behave, check what comes back, and approve anything that touches production.
What I bring
Most of what made this project work happened between the systems — and between the people.
Read the API doc and the contract, prove it in a sandbox, handle what breaks in production, write down how it works.
Goals from the owner became 17 product areas and a phased plan — storefront first.
Readiness checklist, health checks, runbooks — and a database move nobody noticed.
An illustrated operations guide, four role-based staff manuals and 17 operational documents; staff make day-to-day changes themselves.
Sixty-seven releases, each small enough to reverse, each with notes the client could read.
WeChat compliance, entity filings, food-seller registration, domain and HTTPS. Someone has to read the rules and file the forms.
Owner, staff, accountant and vendor engineers — in English and Mandarin, remotely from Australia.
Judgement
Four calls that shaped the product more than any feature did.
Every banner was a request to whoever could code. Days of delay.
Marketing publishes campaigns and landing pages itself.
The incentive scheme was still being negotiated while the platform was built.
Rules are settings in the console — reversible, no release needed.
Data in files. The easy fix was a maintenance window — while dealers were placing orders.
Both stores ran side by side, switched per data type. Nobody noticed.
A payment landing after a cancellation meant a ticket, a manual refund and a note for the accountant.
Detected and refunded through the original channel, with a ledger entry. No human needed.
Also shipped
Sales staff in 20 countries collected customer details on paper. Ten weeks later: a phone form anyone can hand to a customer, a live map for the owner, and a feed that keeps the main platform in step.
Approach
Habits from a year as the only person between an owner's ambitions and a working product.
"We need coupons" usually means "dealers are losing customers to price".
Scope and trade-offs go into documents people can disagree with — and specs AI agents can build against.
Sixty-seven small releases beat one big launch. "Not quite" stays cheap.
Success is when the client stops needing to call me.
Payments reconcile, refunds are traceable, migrations lose nothing.
If a missing form blocks the launch, filing the form is my job too.
Technical foundation
I built all of this myself, directing AI agents, so I can talk to a customer's engineers as a peer, evaluate a vendor on substance, and debug an integration from the logs. The full engineering write-up is one click away.
Claude Code · Cursor · Node.js · PostgreSQL · REST APIs · OAuth · Webhooks · WeChat Mini Program · WeChat Pay · Alipay · AdaPay · Alibaba Cloud · Docker · Nginx · DNS / TLS · GitHub Actions · 760+ automated tests
Contact
Final-semester Master of IT at UNSW, graduating December 2026. Based in Canberra, open to junior roles where AI is how the work gets done — AI solutions engineer, AI implementation consultant or forward deployed engineer — in Canberra, Sydney or remote. I reply to every message, in English or Mandarin.