Skip to content

New Zealand

Ben Gao

Software engineer & consultant

Cloud, integrations, ecommerce

I build and repair the systems a business runs on: the web platform, the seams between it and everything else, and the cloud underneath. Most of my recent work has been ecommerce — Shopify and BigCommerce talking to ERP, point of sale and payments, so stock and money agree — but the same problems turn up wherever a business depends on software it did not write. Lately much of that code is written alongside coding agents, which makes the record of why something was built matter as much as the code itself. Nearly four years of it inside New Zealand retail.

Where I help

Product engineeringAgent engineeringSystems integrationPaymentsCloud & infrastructureReliability & security

Experience

I work across the product and the systems around it: the code, the agents now writing a good share of it, the connections between systems, and the cloud underneath. Ecommerce is the domain I know deepest — not the only one I work in.

Most of it is ordinary, which is the point: a screen someone uses every day, the API under it, the data behind both. Usually TypeScript, React and Node. I like it when the data decides the shape of the screens, so a change stays a small change — and the test that matters is somebody using it without being shown how.

  • Web apps in TypeScript, React and Next.js
  • The API and the data, not just the screens
  • Somebody else uses it before I call it done

I write a lot of code with AI agents now. The interesting part is not the code: a session decides something, the tab closes, and the reasoning is gone, so the next one decides again — sometimes the other way. The fix is dull. The decisions live in the repository, in Markdown next to the code: what we need, what we chose, what is still open. The agent reads that before it writes anything, and the same argument does not happen twice.

  • Notes in the repo, read before the agent writes
  • AI features where the answer has a shape to check
  • Usage capped and queued, mostly to keep the bill predictable

The unglamorous half of a business: stock in three places, orders entered twice, a nightly export quietly holding it together. I wire the systems together instead — Shopify and BigCommerce into ERP, point of sale and the warehouse — so a sale becomes a pick in minutes. Ecommerce is where most of this has been, and the same wiring feeds the month's numbers, which is why reports can read the live systems instead of last night's copy.

  • Shopify and BigCommerce into ERP, PoS and warehouse
  • Orders and stock that agree within minutes
  • Monthly numbers read from the live systems

Payments are where a small mistake turns into a customer complaint, so most of the work is in the boring parts: what happens if the connection drops mid-payment, if the same notification arrives twice, if a refund lands on an order that has moved on. Most of it is Stripe. The aim is one charge per order, a refund where it belongs, and a trail back to the order for whoever does the books.

  • Stripe, end to end
  • One charge per order, even when the network drops
  • Every transaction traceable to its order

Replacing a business-critical system in one go is how these projects go wrong, so I move it across in stages and keep the old one running until the new one has earned its place. Google Cloud, with the setup written as code so it can be rebuilt the same way next time. Costs are what clients ask about first; they follow usage, not the peak someone provisioned for and never hit.

  • Change in stages, no shutdown weekend
  • Google Cloud — I'm a certified Associate Cloud Engineer
  • Costs that follow usage

Business systems do not keep office hours, and neither do the failures. What I set up is unremarkable on purpose: things that restart themselves, changes that go out daily behind checks, and a record of what happened, so an incident gets explained instead of guessed at. Customer data masked, access limited to the people who need it — that part is not optional.

  • Services that restart themselves
  • Changes that ship daily, behind checks
  • Customer data masked, access limited and logged