Hobby projects · Research system · AI

Iran Monitor

A multi-source monitoring and Telegram alert system for regional events.

A language model classifies each incoming signal, removes duplicates, and adds the context that goes into the alert.

Why I built it

My wife was living in Istanbul during the escalation that followed the strikes on Iran. I wanted minutes of warning, not a news cycle.
The whole system answers one question: did the conflict reach Turkey. That question sets the sources, the alert levels, and what counts as urgent.
The design accepts its own limit. No public source can beat a national siren, so the system aims to be early, not first.

I wired up thirteen independent sources, each with its own client, schedule, and failure handling.
Seismic: EMSC over a WebSocket for immediate events, and USGS as the backup feed.
Aviation: adsb.lol and the OpenSky Network for aircraft movement, filtered to military transponders over a Turkish bounding box, plus NOTAM airspace notices for the Turkish, Iranian, Iraqi, and Syrian flight regions.
Civil defense: the Israeli Home Front Command alert feed, as the earliest public signal of an exchange in the region.
News: GDELT for global event coverage, and five RSS feeds from Al Jazeera, the BBC, Google News, The War Zone, and Breaking Defense.
Open sources: selected Telegram channels and Twitter search, scored for credibility and bias before they can raise an alert.
Reference data: ACLED conflict records, UK FCDO travel advice, and NASA FIRMS heat detections.
It normalizes and deduplicates incoming signals before alert processing.

Telegram delivers alerts and health messages.
The analyzer sends each signal to a language model, which classifies it, filters duplicates, and adds context.
The model is set by configuration, so the service can move to another provider without a code change.
The service persists content hashes and Telegram status-message identifiers across restarts.