72 events in one night 

I had a spreadsheet. 72 dance events across Europe. Festivals, weekenders, marathons, retreats. Each one needs a title, dates, venue, description, dance styles, ticket links, poster images, SEO metadata, and a push to Google. Manual entry takes 15 to 20 minutes per event. 72 events is a full work week of copy-pasting.

I don't copy-paste.

The machine

I built a queue. Each event is a row with a status: PENDING, RUNNING, COMPLETED, FAILED. A worker picks up 4 events at a time. For each one, the same chain runs: scrape the source page, extract structured data with AI, validate against business rules, create the event, generate branded posters in multiple formats, push the URL to Google.

Every step can fail without killing the others. A bad poster doesn't block the description. A missing ticket link doesn't stop the venue. The pipeline knows the difference between broken and incomplete.

The number

59 published. 13 rejected.

The 13 weren't bugs. They were events where the source page didn't have enough data. No venue listed. Wrong dates. A broken link. An Instagram profile instead of an event page. The pipeline stopped. It didn't guess. It didn't fill in blanks.

A system that publishes 72 out of 72 by making things up is not automation. It's a liability. The one that publishes 59 and flags 13 for review is the one you can trust with your data.

What it costs

One server, already running. No cloud functions, no managed queues, no Kubernetes. A file, a worker process, and validation rules that know when to say no. Built on PulseAI, a workflow engine I designed for multi-step automation where every decision is traceable and every failure is recoverable.

The whole run took about 30 minutes. Not a week. Not a day. 30 minutes. Then I went through the 13 that failed, fixed what I could, and ran them again.

The hardest part of automation isn't making it go. It's making it stop.