GeoRadX is a diagnostic engine wrapped in a working clinic platform. This is how the pieces fit — from the moment a film is loaded, through the copilot and long-term patient care, to how a whole fleet runs on it.
The core loop runs entirely on the device in front of you. You load a chest X-ray, and GeoRadX returns findings pinned to the regions that produced them — with the measured reasoning behind each one — in seconds. No upload, no wait, no connection.
Runs fully offlineOpen a chest film from the device or the imaging hardware. It's de-identified locally before anything else happens.
The film is measured region by region against the local Master Grid — on the device, no cloud in the loop.
Each finding is placed at the region that produced it, with the measurement that drove it — not a heatmap to guess at.
GeoRadX supports the read; it never replaces it. The clinician reviews the reasoning and makes the call.
The engine produces precise, structured output. The copilot — GeoRX Lite, powered by Gemini — turns that into language a clinician can talk to: what does this region mean, how does it compare to the local population, what should I look at next.
Crucially, it stays grounded in what the engine actually measured. It explains and contextualises; it does not invent. The engine holds the truth, the copilot holds the language.
Internet requiredEvery read a patient receives is kept in their record. Because GeoRadX turns each film into the same structured object, it can compare them region by region across months — showing whether a finding is clearing, holding, or spreading, with numbers rather than a clinician's memory of two films side by side.
This is what turns a one-off screening tool into ongoing care — and it's the data a health programme needs to prove its work.
Internet requiredGeoRadX isn't just a reader — it's the operational layer a clinic or a whole NGO fleet runs on. Each organisation manages its own people, its own devices, and its own data, semi-independently. We provide the secure platform; the organisation runs on it.
An admin authorises who can use GeoRadX under the clinic. We never decide who has access inside your organisation — you do.
Films never leave the device. Only the engine's numeric measurements sync — and they can't be turned back into an X-ray or traced to a patient.
An NGO runs many teams under one organisation. Deployments are managed across sites and borders, each device calibrating to where it is.
Query where you've operated and what you've found over time — the in-organisation research a health agency needs for its own reporting.
GeoRadX runs on usage credit. You buy it in bulk, and every part of the product draws from the same pool — so there are no per-seat fees and no separate add-on bills for standard use.* You decide, day to day, whether a credit goes to a reading or to the copilot.
* Organisations on a Pro account — for running their own custom Master Grid, querying their operational history, or conducting internal research — pay a per-device fee of $80 / device / month. A custom Master Grid, if they choose to set one up, carries a separate maintenance subscription of $300 / month. Standard usage credit is unaffected either way.
Bought in bulk. A reading is a fixed number of credits; the copilot draws from the same balance.
Scans, copilot questions, population comparison, research — all from the same balance, your call.
Contributing readings earns credit back — used as a discount on your next bill, or carried a short while.
This is the distinction that matters most in the field. Inference on a film runs on the device, offline, every time. The layers on top — the copilot, longitudinal tracking, and grid updates — reach our cloud, and need a connection when they run.
Lose connection in the field and you lose the assistant, never the diagnosis. The engine keeps reading on the last grid it received until it can reach the cloud again.
This is how GeoRadX works as a tool. If you want how the engine itself reasons — the architecture that makes it traceable and unable to hallucinate — that's a separate page.
A full step-by-step tutorial will live here once the app is available.