On-premises model serving
Models are served from your organization’s GPU servers.
- Your existing GPU infrastructure is used.
- The models to run are chosen together with your organization.
- The assistant and the clinical skills connect to those models.
FOR PHYSICIAN / FOR HEALTHCARE ORGANIZATIONS
Clinical AI for your physicians, under your organization’s control.
Run For Physician’s assistant for physicians and its clinical skills on models served from your organization’s GPU servers. Use cases, sources and access limits are defined together with your organization.
Physician workspace
Could you calculate the CURB-65 score?
Clinical scoringOrganization guideline
ORGANIZATION NETWORK
Model serviceYour organization’s GPU servers
The platform is For Physician’s. The control is yours.
On-premises model serving
AI models are served from your organization’s GPU servers; the assistant and the clinical skills run on those models.
A workspace made for physicians
Not a general-purpose chat screen: an interface built around the clinical question, the sources and the physician’s workflow.
Task-specific clinical skills
Tools configured for one job, such as prescription review, discharge summary drafting, current literature and clinical scoring.
Two deployment models
One of two paths is chosen to fit your infrastructure; physicians use the same workspace in both.
Managed by For Physician
Models are operated on For Physician’s infrastructure.
- No GPU investment is required.
- Model providers and data boundaries are shared during discovery.
- The scope is still defined together with your organization.
From clinical question to a sourced answer.
The steps a physician takes in the workspace, shown on a fictional case.
Physician workspace
68-year-old patient, suspected community-acquired pneumonia. Alert, respiratory rate 24/min, blood pressure 110/70 mmHg, BUN 26 mg/dL. Could you calculate the CURB-65 score?
SkillClinical scoring
Organization sourcePneumonia guideline · sample document
CURB-652 points · moderate-risk group
- Confusion
- None
- 0
- BUN above 19 mg/dL
- 26 mg/dL
- 1
- Respiratory rate 30/min or above
- 24/min
- 0
- Systolic below 90 or diastolic 60 or below
- 110/70 mmHg
- 0
- Age 65 or above
- 68
- 1
Source: Pneumonia guideline, scoring section
The score supports your assessment; the clinical decision is yours.
Request specialist adviceThe context was opened to a pulmonology specialist.
Fictional example; it contains no real patient data. The screen is a simplified view of the product.
Clinical skills in the workspace
- Assistant
- A general-purpose physician assistant for clinical questions.
- Prescription
- A review of the interactions and checkpoints in a prescription.
- Discharge summary
- A structured discharge summary draft for the physician to edit.
- Current literature
- Source-focused search across selected guidelines and publications.
- Clinical scoring
- Consistent application of defined clinical scores.
Which component runs where?
We show separately what is settled today in on-premises model serving and what will be defined together with your organization.
PhysicianYour organization’s physicians
FOR PHYSICIAN PLATFORM
Physician workspaceWeb and mobile app
Assistant and skill routingPasses the question to the relevant skill and model
ORGANIZATION NETWORK
Model serviceModels running on your organization’s GPU servers
DEFINED TOGETHER IN DISCOVERY
- Identity and sign-in
- Organization sources
- Logs and records
- HIS / EHR connection
Where the components other than the model service run is defined together with your organization during architecture validation.
Governance
In the platform
- Per-physician access: each physician works with their own account.
- Your organization chooses the models and the sources.
- AI output is supportive; the decision stays with the physician.
Defined together in discovery
- Role and permission boundaries
- Log and data retention policy
- The update process for model and skill versions
AI output is supportive; the clinical decision belongs to the physician.
A ready platform, configured for your organization.
Models
Which models are used is chosen together, to fit your infrastructure and your use cases.
Skills
Which clinical skills are enabled is decided by the needs of your physician group.
Sources and rules
The guidelines and documents answers rest on, and the access limits, are defined by your organization.
A pilot in four stages
- 01
Clinical and technical discovery
The physician group, the use case, the data boundaries and the existing infrastructure are clarified.
- 02
Architecture validation
Model serving, authentication and network requirements are reviewed together with your team.
- 03
Limited physician pilot
Use begins in a selected unit with selected skills, on fictional or permitted data.
- 04
Evaluation and the roll-out decision
Usability, answer quality, latency and physician feedback are evaluated together.
Questions decision makers ask.
Which models can run on our existing GPU infrastructure?
The assistant and the clinical skills run on models served from your organization’s GPU servers. Which models are suitable depends on your GPU capacity; this is defined together during clinical and technical discovery.
Does only the model run on premises, or the other components of the application too?
What is settled is that the model service runs on your organization’s servers. Where the workspace and the other components run is addressed during architecture validation, according to your organization’s requirements.
Which systems does patient data pass through?
The answer depends on the chosen architecture, so we do not give a general one. The data flow is mapped component by component during architecture validation, and the conditions for processing personal data are set by your organization’s policies. The pilot runs on fictional or permitted data.
How are sign-in with the organization’s identity (SSO) and HIS / EHR integration handled?
Today physicians sign in to the platform with their own accounts. Your expectations for organizational sign-in and an HIS / EHR connection are heard during discovery; their scope is settled during architecture validation.
Who manages model and skill updates?
In on-premises model serving, your organization decides which models run. The process for updating models and skills is defined together during discovery.
How are usage and answer quality measured?
The criteria are defined together before the pilot starts: usability, answer quality, latency and physician feedback. The evaluation at the end of the pilot uses those criteria.
How do we proceed without GPU infrastructure?
You can proceed with the model managed by For Physician, which needs no GPU investment. Model providers and data boundaries are shared during discovery.
Who from our organization needs to take part in the pilot?
A clinical lead, a contact from information technology and the physician group that will take part in the pilot. Who joins at which stage becomes clear in the first meeting.
FOR PHARMA COMPANIES AND ASSOCIATIONS
Create AI tools for physicians.
Connect them with physicians.
For Physician develops AI tools for specific specialties or physician groups and gives organizations access to an active physician network.
Let us define the clinical and technical scope for your organization together.
The first meeting focuses on the use case, the physician group, the existing infrastructure, the data boundaries and the pilot scope.