I built a real-time OSINT dashboard tracking the Iran-US war's impact on infrastructure rather than military play-by-play.
Stack: Next.js 16 (static SSG on Cloudflare Pages), react-globe.gl with Three.js (14 DRACO-compressed GLB weapon models, 45KB total), Cloudflare Workers for live data every 10-15 min, hand-rolled SVG sparklines, Tailwind CSS 4.
What it tracks: 3 AWS + 1 Microsoft data centers physically struck by Shahed drones, 16 submarine cables (Meta 2Africa declared force majeure), Hormuz at 3% throughput, 48 weapon systems with burn rates and depletion projections, 357 credibility-tiered sources across 6 tiers.
The editorial pipeline runs an 11-phase OSINT scan (Exa neural search, Twitter/bird CLI, Telegram, GDELT) with a 4-test bias detection protocol and Admiralty confidence ratings. Every finding screened against 8 disinformation patterns before entering the timeline.
The interesting technical bits: editorial status overrides when cloud provider APIs report "operational" while facilities are physically offline (ASSESSED badge system), a prebuild script that auto-syncs blog metadata/sitemap/commentary on every deploy, and the whole thing is curated by a persistent AI agent (Claudicle, open-source: github.com/tdimino/claudicle) with its own editorial voice and three-tier memory.
I work at an AI/ML startup. This is a side project. Freely available, no paywall.
tamarru•1h ago
Stack: Next.js 16 (static SSG on Cloudflare Pages), react-globe.gl with Three.js (14 DRACO-compressed GLB weapon models, 45KB total), Cloudflare Workers for live data every 10-15 min, hand-rolled SVG sparklines, Tailwind CSS 4.
What it tracks: 3 AWS + 1 Microsoft data centers physically struck by Shahed drones, 16 submarine cables (Meta 2Africa declared force majeure), Hormuz at 3% throughput, 48 weapon systems with burn rates and depletion projections, 357 credibility-tiered sources across 6 tiers.
The editorial pipeline runs an 11-phase OSINT scan (Exa neural search, Twitter/bird CLI, Telegram, GDELT) with a 4-test bias detection protocol and Admiralty confidence ratings. Every finding screened against 8 disinformation patterns before entering the timeline.
The interesting technical bits: editorial status overrides when cloud provider APIs report "operational" while facilities are physically offline (ASSESSED badge system), a prebuild script that auto-syncs blog metadata/sitemap/commentary on every deploy, and the whole thing is curated by a persistent AI agent (Claudicle, open-source: github.com/tdimino/claudicle) with its own editorial voice and three-tier memory.
I work at an AI/ML startup. This is a side project. Freely available, no paywall.