AI transparency
Osprey uses AI to prepare diagnostic reports. This page explains what the engine does, which models it uses, how a doctor stays in control and where its limits are.
1. What the engine does
Osprey reads a patient intake, searches curated disease cards for the chosen specialty and writes a structured report. Every statement in the report cites the evidence it rests on.
From the same case it can prepare standard guidance, a checklist, an orientation and a scorecard. Each output carries the doctor-review notice.
2. Models and where they run
The engine calls language and embedding models through the Dudoxx model router. In an installation, the models run on the operator's own infrastructure, so patient data stays there.
Patient data from an installation is not used to train models.
3. Human oversight
A qualified doctor reviews every report, edits it where needed, signs it and decides. Osprey makes no diagnosis and no treatment decision on its own. The doctor can reject any statement the engine proposes.
4. Curated knowledge
Disease cards are written and maintained with cited sources. An optional curator can research cards and file cited proposals; it never publishes a change by itself. A person approves each change.
5. Limitations
AI output can be incomplete or wrong. The engine knows only what the intake and the disease cards contain, and the cards cover a limited set of specialties and conditions. It cannot examine a patient.
We publish no accuracy or outcome figures for Osprey. Its output supports the doctor's work and never replaces the doctor's judgement.
6. Regulatory status
The regulatory classification of Osprey: [to be confirmed]. This website makes no claim of regulatory approval.
7. Report an issue
If you notice a wrong, misleading or harmful output, tell us. Write to the email address in the imprint and describe what happened. Do not send patient data by email.
The German version of this page is the legally binding one. This translation is provided for convenience.