v1: Launching WakeUpNeo.ai — A Real-Time Global Risk Monitor
Why we built a 0–10 risk meter that reads the global news cycle in real time, and how the first version works under the hood.
The original idea was simple: could you build a single number that tells you, at a glance, how alarmed the world's press is right now? Not a sentiment dashboard with 40 charts, not a news aggregator you have to read — just a number, updated continuously, with enough transparency that you can understand exactly what's driving it.
The v1 approach
Five RSS feeds (BBC World, Al Jazeera, NPR, Reuters, AP) are polled every 2 minutes using conditional HTTP — if nothing has changed, the server sends a 304 and we skip processing. New articles are deduplicated by content hash and stored in Firestore with a 72-hour TTL.
Each article is scored on four factors: urgency (keyword density), sentiment (negative word density), veracity (static source trust score), and recency (exponential decay with a 6-hour half-life). Claude 3.5 Haiku enhances the top 50 articles per cycle with deeper semantic analysis.
score = 0.40×urgency + 0.30×sentiment + 0.20×veracity + 0.10×recency
The meter value is the mean of the top-100 scoring articles in the rolling 72-hour window, updated live in the browser via a Firestore real-time subscription.
Known limitations from day one
- English-language Western sources only — events with limited English coverage are structurally underweighted.
- Lexicon scoring is gameable by alarmist headlines that don't reflect serious events.
- No market signal — financial stress is invisible to a purely text-based system.
- Source veracity is a static constant, not a per-story credibility assessment.
v2 addresses the market signal gap. The others are on the roadmap.