Founding cohort · 8 seats · starts Thu, 20 Aug

Most people can get an app live. Then it breaks, and they stop.

Every AI-coding tutorial ends at the happy path — the deploy succeeds, the demo works, the video stops. Then a build fails, or a key leaks, or the thing falls over at 2am, and you have no idea which of those is even happening.

This is five evenings on the whole arc: get it built, get it live, get it found, and know what to do when it breaks. Every session ends with your own app in a better state — and half of every session is spent deliberately breaking things, because that's the part nobody teaches.

Five Thursdays

8–10pm SGT, live on Zoom

8 seats

Small enough that I unblock you personally

10 hours, hands-on

You type. I don't demo at you

Bring your own project

Sessions are built around it

The arc

Five sessions, five working outcomes

Pre-work lands five days before each session so the live two hours go on building, not watching slides. The dates skip two Thursdays on purpose — I'm not teaching you anything useful after a 10-hour conference day.

01

Get a machine that can build

Thu, 20 Aug 2026

You leave with: Claude Code running, a repo on GitHub, and a live URL.

  • The eight terminal commands that actually matter
  • Install Node, git and Claude Code — and authenticate it
  • git init → first commit → push, and why a commit is your undo button
  • GitHub → Vercel, first deploy

Then we break it

A failed build. Read the Vercel log, find the one line that matters, fix it, redeploy.

02

Secrets, keys, and the API

Thu, 3 Sept 2026

You leave with: Your live app calls the Anthropic API without leaking a key.

  • Why a key in your source is a financial risk, not a style issue
  • .env.local, .gitignore, and what “works on my machine” really means
  • The three Vercel environments, and why one variable holds three values
  • The Anthropic API: models, tokens, streaming, and where the money goes
  • NEXT_PUBLIC_ — the prefix that publishes your secret to the world

Then we break it

Deploy with a missing env var. It passes locally and fails in production. Diagnose it from the runtime log.

03

Launch it like you mean it

Thu, 17 Sept 2026

You leave with: A real domain, and a site both people and AI can find.

  • Custom domain, DNS, HTTPS
  • The SEO fundamentals that still matter: metadata, sitemap, robots, structured data
  • AEO and GEO — how to be the thing an AI answer cites
  • llms.txt, schema markup, and answering in the opening 50 words
  • Analytics and consent, without being creepy about it

Then we break it

Ship a page with no metadata. Look at how a crawler and an LLM each see it. Fix it. Compare.

04

When it breaks at 2am

Thu, 24 Sept 2026

You leave with: You can diagnose a production failure you didn't cause.

  • Logs, errors and traces answer three different questions
  • Vercel runtime logs — and log drains, when you outgrow them
  • A loop you can repeat: reproduce → isolate → bisect → fix → guard
  • Pointing Claude Code at a failure: feed it the log, not a vibe
  • Error tracking, uptime checks, and alerts you'll actually read

Then we break it

I ship a real bug into a shared app. You diagnose it from the logs alone — nobody tells you what it is.

05

Scale, and the data decisions

Thu, 1 Oct 2026

You leave with: You can justify your storage choices out loud.

  • Neon Postgres: branching, pooled vs direct connections, migrations
  • Supabase: when the bundled auth and storage is worth the coupling, and when it's a trap
  • Upstash Redis: cache, rate limits, queues — and why “serverless Redis” is a different product
  • Choosing between them honestly: a decision table, not a favourite
  • Caching, ISR, and what actually costs money
  • Rate limiting and abuse — your API key is a wallet

Then we break it

Hammer your app until it falls over. Add a cache and a rate limit. Hammer it again.

Founding rate

S$500 S$900

This is the first run, so it's priced for the first run. In exchange I'll ask you for an honest testimonial at the end and permission to say what you built — not a five-star review, an honest one. Later intakes pay S$900.

No account needed. Your seat is held for 72 hours while you pay by PayNow — the amount and reference are locked, so there's nothing to type. Sessions run 8–10pm SGT.

Honest bit

Don't book this if…

  • You want a certificate more than a working app.
  • You can't commit to five Thursday evenings — the sessions build on each other, and session 5 assumes sessions 1 to 4.
  • You're already shipping to production comfortably. This will bore you; email me about the advanced material instead.
  • You want someone to build it for you. That's a different service — see Groundwork.

One more thing worth being straight about: this is a non-accredited corporate workshop. It isn't SSG/WSQ courseware and it isn't SkillsFuture-claimable. I hold the ACLP from the Institute for Adult Learning in a personal capacity, and I've run 20+ GenAI sessions in the last 18 months — including 10 classes for ITE College Central and two four-day cohorts for a manufacturer in Johor. That's the track record; the funding isn't there.

Free · no catch

Not sure yet? Come to a Friday first.

Same material, one third the depth, no hand-holding — I build in public and you watch what actually goes wrong. Free, 8–9.15pm SGT.

Fri, 7 Aug 2026

A GeBiz and news monitor, built live on Telegram + Claude Cowork

Watch GeBiz for tenders worth bidding on, read the news daily, and push both to Telegram and email. Connectors, not code. Built unedited, mistakes included.

Fri, 21 Aug 2026

The keys, the bills, and the things that leak

A real Anthropic spend dashboard, and a real near-miss with a leaked key.

Fri, 28 Aug 2026

How I get AI to recommend my clients' sites

Actual AEO and GEO results from OTG projects — what worked, what did nothing.

All webinar dates

Your seat is held for 72 hours while you pay by PayNow. Full booking, refund and transfer terms are on the policies page.