Development of AI Chatbot for Medical Institutions
Medical chatbot operates under strict constraints: doesn't diagnose, doesn't replace doctor, respects medical confidentiality. Yet solves real problems: appointments, information, patient routing.
Allowed and Prohibited Functions
What bot does:
- Schedule doctor's appointment (specific specialty or doctor)
- Cancel and reschedule appointments
- Procedure preparation (fasting for tests, medication withdrawal)
- Working hours, addresses, contacts
- Test result status
- Specialist routing by symptoms (routing, not diagnosis)
- Appointment reminders, preventive checkups
What bot never does:
- Never diagnoses
- Never recommends specific medications and dosages
- Never interprets test results without doctor
- Never discourages doctor consultation
Symptom Checker as Router
"Find specialist by symptoms" — not diagnosis, but help choosing specialty:
Patient: "Stomach ache"
Bot: "Tell me more:
- Sharp or dull pain?
- How long?
- Any fever?"
→ Based on answers: "You should see gastroenterologist.
Next available appointment — tomorrow at 14:00."
Explicit warning on acute symptoms: "If pain is severe and sudden — call ambulance (103) or ER immediately."
HIPAA/Law 152-FZ and Confidentiality
Medical data is special category of personal data (Law 152-FZ). Infrastructure requirements: data stored only in Russian data centers, encryption at rest and in transit, strict audit log. LLM: only on-premise or Russian clouds (Yandex Cloud, Sber Cloud) without data export.
Integration with Medical Information Systems
Medical Information System (MedWork, EMIAS, QMS): API for doctor schedule, appointment creation, test status checking. FHIR R4 — international standard for medical data interoperability.
Metrics: % calls converted to self-appointment via bot, reduction in call-center load, patient CSAT.