Model foundation
Real-world clinical histories help the model learn medical language, presentations and reasoning patterns across time.
Volume does not prove representative coverage of every population, language, setting or specialty.
Current testing scope
Clinical specialties represented in Diagnify testing.
This page lists only the areas currently represented in testing. Diagnify’s broad clinical training corpus helps adapt the model, but it does not by itself establish specialty performance, safety or authorisation.
Two different questions
A broad training corpus is not specialty validation.
Diagnify currently describes its internal fine-tuning corpus as more than 50 million historical consultation records spanning 50 years, with human feedback during development. The count remains company-supplied pending the full data statement; specialty readiness requires separate evidence.
Specialty validation
Each intended task needs relevant cases, reference standards, subgroup analysis, human-factors testing and monitoring in the target workflow.
Separate evidence and guideline retrieval must also reflect the applicable jurisdiction and current practice.
Separate controls: training-time human feedback improves the model; clinicians remain responsible for reviewing outputs during use.
Listed areas
Testing is currently limited to two clinical contexts.
Each area presents different reasoning, evidence and safety challenges. Results in one context must not be assumed to transfer to another.
General Practice
Status: currently in testing.
The evaluation focus is structured support for undifferentiated presentations: problem representation, red-flag review, differential construction and questions that may clarify the next clinical step.
- Broad presentations with incomplete information
- Explicit attention to important alternatives
- Review against local pathways and continuity of care
Behavioural Paediatrics
Status: currently in testing.
The evaluation focus is structured organisation of developmental and behavioural information, including context, functional impact, alternative explanations and questions that require clinician verification.
- Developmentally informed history structure
- Family, school and social context where relevant
- Clear separation of observations from interpretation
No broader specialty claim: specialties not named on this page are not claimed as supported. The presence of a specialty here means only that it is represented in current testing; it does not establish performance, safety or regulatory status.
Readiness criteria
“In testing” is narrower than “supported.”
A specialty advances only when the data, evidence, evaluation and workflow all fit its intended use.
Relevant data
Confirm representative cases, reference standards, exclusions and clinically important subgroups.
Expert review
Use qualified reviewers, documented correction criteria and adjudication for difficult or disputed cases.
Current evidence
Ground outputs in versioned guidelines that fit the population, jurisdiction and clinical setting.
Task-specific evaluation
Measure meaningful errors, workflow effects, human factors, drift and monitoring requirements.
Private deployment is not automatic validation: a dedicated API environment still requires evaluation for the customer’s task, users, population and setting.
Use during evaluation
Keep every output inside a clinician-led process.
Confirm the context
Check that the case and task fall within the approved local evaluation scope.
Verify the reasoning
Compare suggestions with the history, examination, authoritative evidence and local guidance.
Record disagreement and failure
Capture omissions, irrelevant suggestions, bias concerns and unsafe outputs for evaluation.
Retain responsibility
The clinician remains responsible for assessment, escalation, investigation and management.
Two offerings, one evidence standard
Use Diagnify directly or build in a dedicated environment.
Transparent scope
No unsupported specialty pages.
New areas will be listed only when there is a genuine evaluation scope to describe.