Clinical fine-tune
A selected open-weight base is adapted for medical reasoning. Diagnify does not claim to have originated the base foundation model.
Diagnify private platform · access by review
Dedicated medical model and evidence infrastructure.
Build your medical AI product on a separately hosted Diagnify environment: a clinically fine-tuned language model, a customer-specific evidence-retrieval store and deployment controls defined for your use case.
The foundation
Not a generic chatbot endpoint.
Diagnify selects licensed open-weight base models, clinically fine-tunes them on a governed historical consultation corpus with human feedback, and keeps current guideline retrieval in a separate inference-time system.
Human feedback
Clinical reviewers correct and assess examples during model-development workflows. This is distinct from clinician oversight when the system is used.
Separate evidence service
Guidelines and source documents remain outside the model weights and are retrieved at answer time with source, date and jurisdiction where available.
Corpus description: Diagnify currently describes its internal fine-tuning corpus as derived from 50M+ historical consultations spanning 50 years. This does not mean 50 million unique patients or equal representation across countries, settings, languages, specialties or populations. Review the documentation status.
Private deployment
A model environment for your product—not a shared customer workspace.
Dedicated model endpoint
A separately hosted endpoint is provisioned for the approved organisation and use case, with model version and change controls defined during deployment.
Customer-specific retrieval
A separate retrieval store can combine Diagnify’s maintained evidence sources with approved organisation-specific materials under agreed provenance and update rules.
Evaluation before release
Teams define intended use, evaluation cases, failure thresholds, human oversight and stop conditions before integrating the service into a live workflow.
Versioned operation
Model, retrieval and API changes are handled through agreed versioning, monitoring, rollback and deprecation procedures.
Isolation and data use
Private, with the boundary written down.
Customer prompts, outputs, uploaded documents, retrieval indexes and organisation-specific configurations are isolated from other customer deployments. They are not used to train or serve another customer’s deployment without explicit written authorisation.
| Control | Deployment requirement |
|---|---|
| Tenant boundary | Endpoint, retrieval store, service credentials and customer-specific configuration are scoped to the approved organisation. |
| Training use | Customer prompts, outputs and uploaded data are not used for another customer’s deployment without explicit written authorisation. |
| Access | Authorised Diagnify personnel and named service providers may process data only as described in the applicable agreements. |
| Retention and deletion | Log, backup, retention, deletion and support-access periods are agreed before production use. |
| Hosting | Region, underlying compute isolation, encryption, subprocessors and cross-border transfers are documented for the selected architecture. |
Important: “Dedicated” does not by itself promise a physically exclusive GPU, hardware host or copy of base weights. The exact technical isolation boundary is specified in the deployment agreement.
Access process
Start with the use case. Then design the deployment.
Describe the product and intended users
Tell us the clinical task, geography, users, workflow and whether protected or identifiable health data is in scope.
Define clinical and technical boundaries
Agree intended use, exclusions, model and retrieval requirements, integration method, human review and escalation paths.
Complete governance review
Document security, privacy, licences, regulatory position, evaluation protocol, retention, support and incident responsibilities.
Deploy, validate and monitor
Provision the agreed environment, validate on representative cases, monitor failures and retain a stop-and-rollback path.
Appropriate projects
For governed medical AI development.
- Clinician-facing decision-support workflows
- Evidence retrieval and structured medical reasoning
- Products with defined human oversight and monitoring
- Teams prepared to perform use-case-specific validation
Due diligence
Inspect the foundation before you integrate it.
Private platform access
Tell us what you are building.
Access is reviewed for intended use, safety, data governance and deployment fit. Include your organisation, product, users, geography and expected data types.