Methodology
How livefromeurope works
Hundreds of newsrooms, more than twenty languages, one clear picture of Europe. Here is the pipeline that turns thousands of daily reports into a few dozen stories, and the European AI behind it.
Every story is written by AI and published without human review. Stories are marked “AI-written” and always link to the original reports.
- of news sources
- Hundreds
- languages understood
- 20+
- rolling news window
- 24 h
- of stories linked to sources
- 100%
The pipeline
- 1
Collect
Around the clock we collect only openly available data: the public news feeds that hundreds of outlets across Europe and beyond publish for syndication, plus curated public channels on social media. We don't crawl or scrape any websites directly. Every source is registered with its country, language and political leaning, so each story can show who reported it. Reports arrive in more than twenty languages and stay in their original language.
- Open data only
- Public news feeds
- Social channels
- No crawling
- 20+ languages
- 2
Understand
The latest multilingual embedding models turn every report into a numerical fingerprint of its meaning. A Polish, a French and a German article about the same event land close together, without translating anything first.
- Multilingual embeddings
- Semantic similarity
- 3
Cluster
Reports from the last 24 hours are grouped by meaning: each group of articles about the same event becomes a candidate story. Groups that mix two events are split apart, and people, places and organisations are recognised across languages so the same name always means the same entity.
- Semantic clustering
- Event detection
- Entity recognition
- 4
Curate
Not every cluster is news. Each candidate is scored on significance, urgency, how fast coverage is growing, how many independent sources report it, and its relevance to Europe. An AI editor then picks the day's stories and keeps a balance across countries, so no single country or language dominates.
- Multi-signal ranking
- AI editor
- Country balance
- 5
Give context
A good story needs more than today's headlines. The language model receives the collected reports, the story's earlier developments, and a continuously updated background brief for each country and topic. A story that has been running for days is written as a continuation, not from scratch.
- All collected reports
- Story history
- Living background briefs
- 6
Write
European large language models write each story in English, using only the supplied reports and linking claims back to the original articles. Separate passes write the headline, assign topics and countries, and mark the story as breaking, trending or an update.
- European LLMs
- Grounded in sources
- Linked claims
- 7
Illustrate
A European image model creates an illustration for each story. The images are deliberately symbolic and faceless. They are not photographs of the events.
- Generative imagery
- Symbolic, not photographic
- 8
Publish & connect
Stories go live on the site, in the app and on social media. Related stories are linked into timelines, and all entities feed the News Graph, a live map of who and what is connected in today's news.
- Story timelines
- News Graph
- Web, app & social
Built on European AI
What you can check on every story
- Every source. Each report the story was built from, with a link to the original article.
- Where coverage comes from. Which countries the reporting comes from and how the sources lean politically.
- How the story developed. A timeline of earlier versions and related stories.
- Who and what is involved. People, places and organisations, linked to Wikipedia and the News Graph.
Limits
AI can misread sources, mix up names or miss context, and clustering can occasionally merge or split events in the wrong place. Check the linked original reports before relying on a story.
Corrections
Found an error? Write to livefromeurope@outlook.com.