RF Hunter

A Raspberry Pi 3 + RTL-SDR dongle continuously sweeps 24 MHz–1.7 GHz in 50 kHz bins, comparing each pass against a rolling per-hour statistical baseline. Anything that deviates enough gets a raw IQ capture and a decode attempt against known protocols — rtl_433 (ISM-band devices), multimon-ng (POCSAG/FLEX pagers, etc), and rtl_adsb (aircraft transponders near 1090 MHz). Anything that doesn't match a known decoder ships here, where it's grouped by frequency into clusters — so the same repeat transmitter (a cell carrier, a garage door remote, a fixed telemetry device) isn't reported over and over.

Genuinely new clusters trigger a short automated write-up reasoning about likely band allocation and anything statistically unusual, sent straight to the operator. Nothing here is decoded content — just RF metadata (frequency, bandwidth, signal-to-noise, rough symbol rate) from a passive receiver.

New clusters also get a small waterfall image rendered from the actual capture before the raw IQ is discarded (click a thumbnail to view it full-size) — Claude looks at that image, not just the numbers, when deciding whether a signal looks structured/bursty enough to flag worth a manual look.

Formerly this hardware ran a TPMS/weather-station sniffer, and before that a from-scratch Random Forest RF classifier — both retired in favor of this general-purpose anomaly hunter.