How our AI reads two hundred sources before breakfast
A look inside the pipeline that turns a continent of headlines into one calm digest every thirty minutes.
Every thirty minutes, while most of Europe is still queueing for coffee, our pipeline has already done a full day''s worth of reading. It has crawled the wire services, the national broadcasters and the regional dailies, stripped the boilerplate, and asked a simple question of every story: is this the same event we have already seen, or something new?
From two hundred tabs to one page
The honest answer to "how do you follow everything?" used to be: you don''t. A human editor skims, guesses and hopes. Our approach is different — embeddings turn every article into a point in space, and stories that describe the same event end up close together. What you see on the front page is not two hundred links but a handful of clusters, each one an event rather than an article.
Where the humans come in
Automation does the reading; it does not do the judgement. Editors tune which sources we trust, how categories are weighted, and they can rewrite or pull any item at any moment. The machine drafts, the newsroom decides.
Why this matters
Speed is not the point — calm is. When every outlet pushes the same alert five times an hour, the scarce resource is perspective. A digest that says "these five things actually happened since you last looked" is our answer to alert fatigue.