03 / Build · Vibe coding

I don't file a ticket and wait a quarter.

The distance between "we should try this" and "it's live" used to be a roadmap, three meetings and someone else's sprint. With AI in the loop it's a weekend — provided the person holding the tools knows what to build and what good looks like. That's the part that doesn't come from the model.

Give me the problem and the tools and I come back with the thing working.

01

What this is worth to you

Not "I know how to code." I don't claim to be an engineer. The claim is narrower and more useful: I can take an idea to something real and measurable without occupying anyone else's roadmap.

Validate in daysA landing page, a working prototype and live traffic against it inside a week — before anyone commits a sprint to the idea.
No handoff taxThe person who designed it, built it and bought the traffic is one person. Nothing is lost being explained three times.
Measurement built inA marketer builds it, so the events, the tracking and the questions week two will ask are in from the start — not retrofitted.
Cheaper experimentsWhen a test costs a weekend instead of a headcount quarter, you can afford to be wrong — which is how you find the things that work.
02

The method

Vibe coding fails when it's just prompting until something renders. It works when it's a process with the same discipline any other build has. This is mine.

01

Decide what must be true

Before anything is designed: what is this for, who is it for, and what single number tells me it worked? Most failed projects were never wrong technically — they were built without a definition of done. If I can't write the success metric in one line, I'm not ready to build.

02

Design it first, properly

Real screens in Figma before a line of code. This is where four years of UX earns its keep — AI amplifies whatever clarity you bring it. Give it a vague idea and you get vague software fast; give it a resolved design and you get that design, working.

03

Write the spec, then scaffold

Data model, page structure, naming conventions and states decided by me up front and written down. The model is extraordinary at execution and indifferent to architecture — so the architecture is my job. Skip this and you get five ways to do the same thing by day three.

04

Build in small, verified loops

One capability at a time, checked in the browser before the next one starts. I don't accept code I can't read back and explain. If I don't understand it, it isn't done — because in three weeks I'm the one debugging it.

05

Instrument before launch

Events, GA4, the funnel steps I'll want to compare. This is the step almost everyone skips and then regrets, and it's the one my background makes automatic — I've never shipped anything I couldn't measure.

06

Ship, read the data, cut

Live, in front of real traffic, fast. Then the marketer takes over from the builder: what actually got used, where people fell out, what to remove. Shipping is the start of the project, not the end of it.

03

"Scalable" is a discipline,

not a framework.

The fair criticism of AI-built software is that it works on Tuesday and is unmaintainable by Friday. That's a process failure, not a tooling one. These are the rules I hold to — and the honest limits of what I'd take on alone.

01Data model before pixelsWhat the objects are and how they relate gets decided once, by me, in writing. Everything downstream is cheap to change; this isn't.
02One way to do each thingOne naming convention, one state pattern, one place styles live. Conventions are what make a codebase survivable by the next person.
03No dependency sprawlEvery library is a future maintenance bill. This site runs on zero — plain HTML, CSS and JavaScript, because that's genuinely all it needed.
04Read every lineGenerated code I can't explain gets rewritten until I can. That rule is the difference between building software and gambling.
05Small blast radiusShip narrow, behind a flag or a subdomain, to a slice of traffic. Nothing goes live in a way that can take something else down with it.
06Performance and access are designLoad time, keyboard paths, contrast, reduced motion. They get decided in Figma, not patched after someone complains.
07Know the handoff pointAuth, payments, personal data, anything at real scale — that's where an engineer takes over, and I hand across a documented codebase rather than a mystery.
08Know what I'm notI'm not a backend engineer and I don't pretend to be. I'm the person who gets you to a working, measured version one so the engineers build the right thing.
04

Built this way

Designed, built and deployed by me — no template, no team, no agency. Starting with the page you're reading.

This portfolio2026 · Live

Three pages, hand-built. No framework, no build step, no dependencies beyond a webfont — custom cursor, scroll choreography, theming and a case-study overlay system, all written from scratch.

HTML · CSS · JS

Exact Fit Technical Services site2025 · Live

Company website built and maintained on a custom WordPress theme — the same site the paid social test pointed its traffic at.

WordPress

Real-estate CRM2025 · Shipped

Designed and shipped the lead capture, routing, cadence and pipeline system at FLIVV — the project that turned ad spend into a revenue number I could actually report.

Product

ChaiWala menu site2026 · Live

A static menu site with content split into a single editable data file, so a non-technical owner can change prices without touching markup.

Static site

yahiyapasha.com2026 · Live

A content site designed and deployed end to end — brief to domain, without a developer in the loop.

Next.js · Vercel

Build stack

Claude CodeFigmaHTML / CSS / JSFramer WebflowWordPressNext.jsVercel NetlifyGA4

Designer. Marketer. Builder. Hire the whole loop.

Most teams buy those three separately and spend the difference on coordination. I'd rather be the person who closes the loop.

Get in touch See the projects