Flevoland Vooruit
A regional newsroom that writes itself — 24/7, from source to article, image and social post.
Flevoland Vooruit reads ~28 regional sources plus the open council records of six municipalities, and writes its own articles from them — with generated imagery, editorial review and social posts. Meanwhile the Afsprakenmeter measures whether political parties do what they promised. One person runs the entire AI newsroom from a single control room.
28
Sources, read continuously
10
Cron jobs as the heartbeat
7
Journalist personas
47 → 87
Lighthouse mobile after the rebuild
What it does
Autonomous news pipeline
Seven to nine LLM steps per article: preselection, relevance, editorial line, deduplication, web research, rewriting, final review and publication — repeating every thirty minutes.
Seven journalist personas
From political analyst to business desk: the model picks the right writing style per topic. Prompts are tunable live in the Prompt Lab.
Afsprakenmeter
Election manifestos of six municipalities linked to actual voting behaviour from council records: a weighted promise-keeping score per party. Political accountability with data instead of opinions.
Flevoland Live
Emergency calls, traffic, bridge openings and weather on one map — geocoded to street level, with privacy protection on ambulance runs.
AI imagery per article
Every publication gets automatically generated imagery (±$0.04 apiece) with category-dependent lighting mood, plus a stock-photo fallback and a media library.
Social on autopilot
Facebook and Instagram posts on fixed slots, comments classified and answered automatically — with community memory that recognises regulars and adjusts its tone.
Machine-readable
Its own MCP server, llms.txt and a token-protected news API: the platform is a reliable source for AI agents too.
How we built it
Next.js 16 on Vercel with Firestore as the data layer; ten cron jobs form the heartbeat, from emergency feeds every two minutes to the daily council scan. An SSR rebuild lifted Lighthouse mobile from 47 to 87.
Cost control is an explicit architecture driver: layered LLM routing with Gemini first and gateways as fallback, conditional research steps and token logging per article — the entire newsroom runs for a few tens of euros per month.
The control room shows every AI step per article including token usage and cost in dollars — full transparency about what the machine did and why. Source management, agenda desk and the social dashboard live in the same cockpit.
Why we built it
Regional news has been hollowed out economically — but the information is there: council documents, press releases, event calendars. The question was whether one person with AI can run a full newsroom that becomes the central source for a province.
The editorial line is a deliberate counter to clickbait: business-like, positive and factual, with incidents as a small service instead of the front page. And with the Afsprakenmeter, timed for the 2026 municipal elections, the platform holds politics to account with data.
Need something like this in your company?