The World Now

The Problem

World news is fragmented, slow, and hard to see spatially. I wanted a single real-time view of global conflicts, disasters, and markets — published faster and far cheaper than a traditional newsroom.

What I Built

A solo, end-to-end product: a Next.js + Mapbox front end on Vercel, an AWS Lambda fleet that scrapes, classifies, and generates articles around the clock, a Catalyst market-prediction engine, MongoDB Atlas, Stripe subscriptions, and a hand-rolled ISR-safe A/B testing layer in ~80 lines of middleware.

Stack

What it is

The World Now is a live AI news platform at the-world-now.com that tracks global conflicts, disasters, and markets on an interactive 3D globe. An autonomous pipeline researches, writes, fact-anchors, and publishes articles around the clock — over 6,000 published as of mid-2026 — with live event feeds and a market-prediction engine called Catalyst.

Architecture

A Next.js front end with Mapbox globe rendering runs on Vercel with ISR. Behind it, an AWS Lambda fleet (Node 20, us-east-1) scrapes sources, classifies events, generates articles via LLM with schema-first JSON output, and enforces mandatory source provenance on every claim. MongoDB Atlas stores the corpus; Stripe powers subscriptions; Resend handles briefing emails. A hand-rolled, ISR-safe A/B testing layer lives in roughly 80 lines of middleware.

Results and lessons

Ahrefs Domain Rating grew 0→15 in about three months on a Bing-first indexing strategy. Prompt caching cut generation cost ~78% to roughly $0.035 per article. The biggest lesson: autonomous content systems need provenance enforcement and per-row prompt/model versioning from day one, or quality drifts silently.

Outcomes

• ✓6,000+ articles published autonomously
• ✓Ahrefs Domain Rating 0 → 15 in ~3 months
• ✓AI pipeline cost cut ~78% (≈$0.035 per article)
• ✓Schema-first generation with per-row prompt/model versioning and mandatory source provenance