Skip to content

Landseed AI · CTDAMS

Every instrument
in the landscape
reports here.

Detections are classified on the device, routed to whoever can act on them in about thirty seconds, and kept as the record that occupancy, abundance and the quarterly report are built from.

The console

This is what an operator sees.

CTDAMS SAMPLE DATA 18:45
Occupancy · Virunga National Parksources
◆ live 3D terrain · real DEM + Sentinel-2 satellite · Contains modified Copernicus Sentinel data 2025 · boundary & Lake Edward geometry © OpenStreetMap contributors (ODbL)drag to orbit · toggle layers · tap to inspect
Networklive
18units online
1,340camera-nights
43alerts · 24 h
Population metrics24 h30 d12 m
Detections by species · 30 d
Detections · last 24 h
Scientific surface · Virunga NPsources
Single-season occupancyMacKenzie 2002 · 1,340 cam-nights

ψ̂ detection-corrected · elusive species carry the largest correction · 95% CI whiskers. Sample data.

Species accumulationrarefaction → Chao2 asymptote
Diel activity · detection kernels
Acoustic array · 30 d
412Leopard calls · array northtriangulated
1,048Bio-acoustic events · full surveylogged
36Gunshot signatures · sector Gops room
7Unknown · queued for analyst reviewreview

Roadmap: monocular distance · acoustic triangulation.

Virunga NP · report librarysources

Every document written automatically on the programme's cadence.

Notifications

The alert arrives where the ranger already is.

There is no app to install and no dashboard to keep open. Alerts arrive in Telegram, on the phone in someone's pocket, because that is what field teams already use and what works on a 2G bar of signal at the edge of a park.

Each message carries what was seen, which unit saw it, where, and how confident the model was. Photographs are attached only when the link can afford them — otherwise the text alone is enough to act on.

Routing is set per class and per station by the people running the landscape: a leopard is filed, a gunshot signature is escalated, an elephant near a village goes straight to the village protection unit.

Boundaries

What it will not do.

Each of these is a decision, not a gap in the roadmap.

Identify a person
The detector knows human. It does not run face recognition and we will not add it — that turns a monitoring system into a surveillance system, which is a different product with different consequences for the people living alongside these parks.
Promise a response time
The response is people in vehicles on bad roads, and we do not control that. What the system controls is when they find out: seconds, rather than at the end of the month.
Hide its uncertainty
Every detection carries its confidence, and low-confidence events go to a review queue rather than to somebody's phone. A gunshot signature at 0.71 is a question, not an alarm.
Hold your data
Readings from your land are yours, exportable in the formats agencies and funders actually accept. If you stop paying us, you leave with the record.

How it's sold.

A subscription, billed annually, in two shapes: hosted, for a single landscape with a handful of stations; and enterprise, for multiple landscapes with your own tenancy and integration into whatever reporting stack you already run.

It scales with the number of stations you operate, not the number of detections they produce — charging per detection would give us an incentive we don't want. Tell us the shape of your deployment and we'll tell you where you land.

The instruments open the door. The intelligence is the business.