A region's events scattered across hundreds of websites

The problem

A region’s cultural events are scattered across public websites, local press, social media and posters. Gathering them by hand, one town at a time, cannot be kept up for long. And when the sorting was manual, only 20% to 50% of what was read was any use for the guide.

What we did

At first the shared calendar was fed by a simple script and by sorting done by hand in a chat with an AI. Now a system collects, every day, what comes out on websites, news feeds, PDFs and social media.

AI reads it in steps: a simpler model goes first, and the stronger one only sees what is left unresolved. Then the dates are checked and duplicates removed, and a morning email shows what came in. One rule is fixed: the system never invents dates, times or performers. Anything in doubt goes to a person.

Organisers can also submit their own events in a dedicated area or through an AI connector.

The result

• 1,141 events from 336 sources in the calendar, collected without anyone combing websites by hand.
• In a test, the simpler model found the same events as the stronger one, with no invented dates.

The same design works for any team that has to keep watch on many sources, whether competitors, tenders, press or prices.