How I work

I take an idea to working software, and decide what is worth building. Since March I have built WLTH's products as well as designed them: AI doing the typing, engineers deciding what goes live.

Build the real thing, not a picture of it. The shared foundation first, as working software rather than a style guide.

I design against the live database most of the time, so the work is hierarchy: what matters first, what can wait, what is noise. A change is briefed, built by the AI, reviewed on a live link, checked by an engineer.

The calls the AI cannot make, twelve-pixel corners, no side menu, take as long as ever. One brief per product keeps us consistent: I read it, the engineers read it, the AI reads it every session.

Twelve lender cards were made in Canva; a rate change ran a 39-task checklist. Nobody asked for a tool; I saw a clunky chain with no source of truth and fixed it. The card became data: it renders from the rates, each change goes for approval, publishing is one verb, brokers see the current card. Twelve went live on 3 July; by October every card is generated and a rate change is three actions.

Six products on one foundation, five of them mine. On the staff portal an automated accessibility check went from 132 findings to none. And no handoff, no spec out of date by the time it is read.

What I got wrong: I let the products drift from the foundation, and one went live missing a piece. The fix was numbered versions. What I said no to: the staff portal began as an intranet; at six weeks 20 of 58 pages were empty, so I hid them, then removed them. Building got fast. Knowing what to build, when to change direction and how to leave it maintainable did not. That is the job.