What We Learned Building an AI Resume Builder
Lessons from shipping CVeezi: prompt design, guardrails and why the boring parts of an AI product matter most.
Every redesign starts with good intentions and a long list of screens. Somewhere around the third page, someone asks which grey the borders should be, and the answer turns out to be different on every screen. This is the moment a design system pays for itself.
Redesigns stall on repeated decisions
Most of the time spent on a redesign is not spent designing. It is spent re-deciding the same things: button sizes, spacing between sections, how a card looks when it is selected. Each decision is small. Together they slow everyone down and make the finished site feel slightly inconsistent.
Writing those decisions down once, as tokens and components, turns them into defaults. Designers stop guessing, developers stop measuring, and reviews move from opinions to a shared reference.
A design system is not a deliverable. It is the agreement that keeps a team from deciding the same thing twice.
What goes in a starter system
You do not need a hundred components. For most business websites, a small foundation covers nearly everything:
- Colour and type tokens named by role, not by value
- Spacing and radius scales that every component shares
- Five to eight core components: button, card, input, navigation and footer
- Clear rules for light, dark and brand surfaces
How we introduce it on a project
We build the system alongside the first two or three key screens, not before them. That keeps it grounded in real needs and avoids a long design phase with nothing to show.
Each new screen then borrows from the system, and anything missing gets added once and reused. By the time the last page is designed, most of it is assembly.
Where to start
Audit what you already have, pick the ten decisions you make most often, and write those down first. Colours, type, spacing and a button are enough to begin. Add the rest only when a real screen asks for it.
