Training tailored to each person, with expert approval
The AI in ATP MedAI finds each person's knowledge gaps, gets them practicing and gives access to your hospital's documents with sources cited. Every suggestion comes with a reason; every piece of content is approved.
Where does the AI work?
Six components, the same principles: reasons, sources and expert approval.
Personal learning plan
Weekly plan and spaced repetition based on role, department, exam results and knowledge gaps.
DetailsCase simulation
A virtual patient role-played from a synthetic case card; a second model scores with a rubric.
DetailsProtocol assistant
Answers only from your hospital's documents, with the source shown; declines out-of-scope questions.
DetailsContent studio
Draft lesson summary, video, question bank and translation from a procedure; nothing is published without approval.
DetailsRisk prediction
Flags staff likely to fail a certification exam or let a certificate lapse, with a person-specific confidence level.
DetailsThree sample flows
Screens are illustrative; cases and documents are examples.
The virtual patient talks, the rubric scores
The virtual patient plays a role using the information on a synthetic case card; the staff member asks questions and makes decisions. When the conversation ends, a second model scores it against a rubric and states plainly what went well and what was missing.
- History-taking, triage, medication safety, patient communication, difficult conversations
- Score and feedback per criterion
- Weak criteria are added to the learning plan
- You correctly asked about where the pain spreads and the accompanying sweating.
- Remember to ask for vital signs before making the triage decision.
Scoring is done by a second model against a defined rubric.
Answers that show their source
The assistant searches the procedures and protocols your hospital uploads and shows which document and section the answer comes from. If the documents don't cover a question, it doesn't answer and refers the user to the responsible person.
- Hospital documents only
- A source with every answer
- Makes no clinical decisions
A heads-up before the exam, before expiry
The model flags at-risk staff based on exam results and training history; each flag comes with a confidence level and a reason. The training lead makes the decision.
- Person-specific confidence level
- Flags with reasons
- Suggested review lessons can be added to the plan
| Staff | Department | Topic | Confidence | Status |
|---|---|---|---|---|
| Nurse | Emergency | Triage exam | 82% | Review assigned |
| Technician | Radiology | Radiation safety | 74% | Expiring |
| Midwife | Delivery Room | Blood transfusion | 68% | Added to plan |
| Care assistant | Internal Medicine | Hand hygiene | 61% | Monitoring |
Confidence is the model's estimate; the decision rests with the training lead.
Six principles for trustworthy AI
AI assists; decisions and approval stay with the experts.
Explainability
Every suggestion is shown with its reasoning.
Safe fallback
If the model can't produce a result, the system falls back to rule-based suggestions.
Relative assessment
Strengths and weaknesses are judged against the organization's average on each topic.
Expert approval
All AI-generated content is approved by an expert.
Source citation
The assistant cites its source in every answer and declines out-of-scope questions.
Predictions with confidence
Risk predictions come with a person-specific confidence level.
Models on our own servers, no patient data
The language models run on our own servers, with no data sent to an external AI service. In the Enterprise plan, the models can also run on your hospital's own servers.
- Case simulation with synthetic case cards
- Protocol assistant uses only documents your hospital uploads
- Only staff, role and department information from the HIS
- Role-based access and audit trail
Sample screen · illustrative data
Infrastructure already at work in the field
ATP MedAI uses the AI infrastructure that runs the ATP BESYO and ATP Özel Eğitim platforms.
Success predictions are recalculated every night for more than 4,000 users.
Live classes on our infrastructure run with 400 to 1,000 concurrent participants.
Accuracy of the AI's first suggestion in question tagging, in an expert-approved workflow.
Source: ATP Eğitim Teknolojileri, our own platform data (August 2026). The figures come from exam preparation platforms, not from hospital use.
AI questions
The limits and safeguards of the AI.
Write to our team with questions specific to your organization.
What if the AI makes a wrong recommendation?
Every recommendation is shown with its reasoning; the training lead sees why before accepting it. If the model can't produce a result, the system falls back to rule-based recommendations. All AI-generated content is approved by an expert before it goes live.
Does the protocol assistant answer from the internet?
No. It answers only from the documents your hospital uploads and cites the source in every answer. It doesn't answer questions the documents don't cover.
Where do the language models run?
The language models run on our own servers, with no data sent to an external AI service. In the Enterprise plan, the models can also run on your hospital's own servers.
Does AI-generated content go straight to staff?
No. Lessons, videos and questions produced in the content studio don't go live until the training lead approves them.
Does the protocol assistant make clinical decisions?
No. It is a tool for training and access to information; it doesn't suggest diagnoses or treatments. Clinical decisions rest with physicians and the care team.
See training and compliance on one platform
A live walkthrough with screens suited to your type of facility; the quote is based on your headcount.
Set up by our team · No patient data needed · Each hospital in its own space