Back to projects

Project detail

AI Website Builder — Local Business Demo Pipeline

2026 — ongoing (inferred)

Lots of great local businesses, high ratings, loyal customers, have a terrible website or none at all. And they know it. But every pitch they've ever heard is "pay me and I'll build you one", which is an easy pitch to ignore.

This pipeline flips the order: build first, talk second. Before any business is contacted, I study the best websites in their industry (what sections they have, what makes them feel trustworthy) and distill it into a blueprint, so every site afterward starts from that blueprint instead of a blank page. Then I look for the right businesses: well-reviewed places with plenty of customers but no website, or a clearly broken one. About half of any raw list gets cut after a person double-checks it. The demo itself is built from the business's own material: their photos, their services, their customers' actual reviews, pulled from what's already public. Nothing on the page is made up. And before anyone outside sees it, every page gets gone over by hand: phone numbers and buttons work, reviews are real and credited, it looks right on a phone screen, it loads fast.

The result changes the conversation entirely. Instead of "would you like a website?", it's "here's your website, want it?" Nothing is ever sent to a business without my explicit sign-off, and if the owner says yes, the same process carries through to the real launch: hosting, their domain, contact forms, and a short lesson in how to run it.

Under the hood

The pipeline is five phases, each with a written output so it's repeatable across industries:

  • Vertical research (once per industry). Audit 5–8 strong competitors (page structure, hero patterns, CTAs, trust signals, mobile UX) and distill it into a template spec: the canonical sections and design tokens every site in that vertical needs.
  • Prospect discovery. Google Maps search filtered to businesses with 4.5+ stars, 20+ reviews, and no site or a broken one. About half of any raw list fails manual verification, so every prospect carries a Maps link for checking.
  • Content gathering. Everything public: the Maps listing, every Google review with attribution, photos, posts, existing site content, social media. For redesigns, the existing site is crawled page by page and their real photos and bios are reused, because the before/after is the pitch.
  • Build. The template spec plus the business's real material goes into a self-contained, responsive HTML demo through an LLM-assisted editor. Nothing on the page is invented.
  • Quality gate. Checked against the spec before anyone sees it: all sections present, click-to-call correct, reviews real and attributed, responsive at 375/768/1440px, loads under 3 seconds, zero placeholder content.

Outreach is a hard gate; nothing is sent unless I explicitly decide. Demos are deployed to Vercel or GitHub Pages only when ready to share; internal previews run over a private Tailscale-served port.

HTML/CSS
LLM-assisted content editing
Google Maps research
Google Reviews
Vercel
DADaksh Adhikari