WHO IT'S FOR
The port operations room — the watch officer whose job is catching the one ship in two hundred that needs a human.
WHY THE AI
Without it, this page is just hundreds of moving dots. The ML service watches every ship every second and catches the instant one behaves abnormally. The LLM service instantly writes the response — the advisory a port officer would draft under pressure. You only look when something needs your attention.
HOW DETECTION WORKS
An Isolation Forest — an unsupervised model that learns "normal" from the live data itself. Each report is scored on speed deviation vs. the ship's own baseline, abrupt braking, and dead stops. Workboats (tugs, pilots) are exempt from speed alerts; alerts need consecutive anomalous reports. Retrains on real data via the RETRAIN ANOMALY MODEL button.
WHY IT MATTERS
A ship stuck where it shouldn't be blocks a channel, delays every ship behind it, and costs a port thousands per hour — catching it in seconds beats finding out late. All on-premise; no data leaves the building.
READING THE MAP
Green = moving normally · gray = docked/anchored (sitting still is normal) · flashing red = trouble · faint line = recent path. Click a ship for details & history, click an alert to zoom to it.
HIDE FERRIES & WORKBOATS filters out ferries, tugs and pilot boats — they stop and start constantly as part of normal scheduled work, so they add clutter, not signal.
Live ships
aisstream.io — real AIS positions
▶
Adds context
ais-ingest · usual speed + area
▶
Detects anomalies
ml-engine · Isolation Forest
▶
Alert raised
only brand-new incidents
▶
Drafts response
agent-worker · local LLM (Ollama)
▶
What to do
advisory in Alerts panel →