August 5, 2026

Overview

SearchBlueprint is a Singapore client's AI search platform for small businesses and agencies. A user enters a website URL. The app builds an SEO and AI-search strategy, a keyword map and a content calendar. Then it writes the articles on the calendar. Site audits, page writing and CMS publishing are built and wait for an October launch.

My role

Core contributor since August 2026, in a small team that the product owner leads. The product vision, pricing and most marketing pages are the client's.

• Over 1,100 commits, second only to the product owner
• Commits on 41 of the 48 days from August 9 to September 25
• 1,049 files added in that window, 499 of them test files

What I built

Live for customers:

• Content Calendar autopilot: it writes each topic at midnight on its date and has a keyboard-friendly month grid.
• Credits system: prices on every Run button, runs recorded at list price, receipts and an admin pricing editor.
• Connect walkthroughs: an animated player that shows each publishing setup click by click.
• Article score panel, built with teammates: bars for six weighted categories, and failing checks ordered by points lost.

Built and waiting for launch:

• Audit Programme: nine recurring SEO and AI-visibility audits on a calendar. I built the engine, runners, report view and tab.
• Publishing integrations: WordPress blocks and pages, Webflow, and one-click GitHub publishing for Lovable and TanStack Start sites.
• Calendar publishing: one-click publish and an auto-publish switch in the Content Calendar.
• Super Page builder and Page Editor: ten page types, each checked against its own content rules. A build becomes an editable page design that publishes as a WordPress page. I made about half the commits on both.
• SearchBlueprint in Claude: a connector that lets Claude use the product, with OAuth 2.1 sign-in and 290 tools. Paid or publishing actions wait for the user's approval.
• Admin tooling: email sequences with on and off switches, a funnel view, and payment and credit-request records.

Quality checks:

• Tap-target, contrast and layout-shift tests.
• A check that stops an agent commit from reverting a teammate's recent work.

How I built it

I build with Claude Code. From August 9 to September 25, 702 of my commits credit Claude as co-author.

• I plan the work and keep the team task queue. Each item records the commit that closed it.
• I ran at least 107 multi-agent workflows from August 26 to September 25, 2026.
• Four Codex sessions ran alongside the credit wallet work.
• My agent rules: one task, one commit, push to main. The build must pass before every commit.
• "Done" means the code ran and I read the output. Unverified work stays in review.
• 85 of my commit messages cite browser, Playwright or screenshot checks.
• A /hunt command runs an audit loop: capture, detect, verify, dedupe and file. It has filed 163 findings.
• Agents never re-run paid DataForSEO or Google calls to check work. They read stored results.

Tech stack

• Web: Next.js 16 (App Router), React 19, TypeScript, Tailwind CSS v4, Puck page editor
• Backend: Supabase (Postgres, RLS, realtime), Stripe, Vercel
• Data and AI: Anthropic SDK, DataForSEO API, Google Sheets, Slides and Search Console APIs, Model Context Protocol
• Quality: Vitest, Playwright, Sentry
• Agents: Claude Code, Codex