The system a real business runs on — designed, built and operated by me

The problem

A growing catering business was managing orders manually — which works until the day it doesn’t. Orders have a lifecycle: placed, confirmed, prepared, delivered, paid. When that lifecycle lives in someone’s head, every busy day is a risk, and questions like how did this month actually go have no reliable answer.

My role

All of them — which is the point of this case study. I did the discovery with the owner, designed the order lifecycle and counting rules, specced the system, built it with AI-assisted development, deployed it, and operate it in production. When something breaks, there is no engineering team to hand it to — I investigate and fix it myself, AI-assisted, and ship regular feature updates as the client’s requests come in.

Decisions that mattered

• The order lifecycle was designed around how the kitchen actually works, not around a textbook state machine — statuses the staff wouldn’t use were cut.
• Counting rules were locked early: what counts as an order, when revenue counts, how cancellations are treated — because a reporting system that can’t be trusted on definitions can’t be trusted at all.
• Built multi-tenant from the start, so a second business can onboard without a rebuild.

What shipped

A production order-management system the business uses every day: order intake and tracking through the full lifecycle, with the reporting layer answering the questions the owner actually asks.

What I learned

Operating a system I built with AI assistance taught me the gap between shipping code and owning it. It pushed me into deliberately building the code-reading depth to stand behind every line in production — the most useful uncomfortable lesson of the project.