Building OptiDesk. Replacing my dad's carbon-copy billing pad with an app, one entry at a time.

Forty-one screens, five judges

Before writing production code, every screen went through a review panel I built for myself: five expert lenses (UX, product, business ops, SaaS patterns, UX writing), each scoring 0 to 10. Nothing passed under an average of 9.5, and one open high-priority issue meant automatic failure regardless of score. All 41 screens went through it. Some took five rounds.

The bar caught things I would have shipped without blinking: an invoice flow that assumed customers have email addresses (most walk-ins do not), money amounts styled three different ways, a delete action one tap from data loss.

Three of the 41: the billing desk, a new invoice, a new customer.

The design system is a lint rule

One person cannot hold a design system in their head across 41 screens and months of changes. So the system does not live in my head. Every color, spacing value and radius is a named token, and custom lint rules fail the build if a screen uses a raw hex color or an off-grid gap. When a padding value needed to change app-wide, it changed in one line.

Strategy and a pitch site in one day

Positioning came before features: "the paper pad, upgraded, not replaced by a computer." Shop owners do not want software. They want their pad, minus the drawer of paper. That one sentence decided what OptiDesk refuses to be: no GST module, no inventory, no login screens. By midnight the pitch site was live, with a pilot funnel and pricing strategy written down.

Three rounds of QA, forty-four findings

I tested the app the way a shopkeeper would break it: typing a phone number into the name field, backing out mid-bill, restoring a backup over live data, killing the app during a save. Three full review rounds produced 44 numbered findings. Every one got a one-commit fix and a re-test before it counted as closed.

Honest entry: one fix got shipped, then reverted, then redone differently. A cosmetic fix behind the main billing button quietly re-opened an older bug where tappable chips got covered. On-device testing caught it. The final version keeps the clean look and guarantees nothing tappable can ever be trapped under the button.

Three of the 44, verbatim from the findings ledger.

Filming an app that lives on a counter

A demo video for shop owners cannot be a screen recording with a mouse cursor. I recorded the real app on an Android emulator with scripted, deterministic taps: seeded shop data so the charts look alive, a cleaned status bar, visible tap indicators. Then the surprises: the emulator records video that plays back 1.5x too fast, and dismissing the keyboard silently eats the next tap, which ruined take after take until I found the cause.

The real thing: a walk-in billed and the PDF shared to WhatsApp, unedited app footage.

A demo you can send on WhatsApp

Optician shops share things one way: WhatsApp. So the demo page treats video timestamps as URLs. Every question in the FAQ answers with a link that opens the demo at the exact second that answers it, like optidesk.app/demo#order:16. One shop owner can forward another a link that says "see, this is how the order tracking works" and it just plays that moment.

Next: real bills

The pilot build is on the counter at Royal Optician, the shop this whole thing was built inside. The next entry is not mine to write at a desk. It gets written by whether my dad and his staff reach for the phone or reach for the pad when the shop gets busy. Activation, weekly bills, time per bill: those numbers will fill this space.

This entry is intentionally unfinished.