Initum AZ - AI Care Assistants

Start 01/01/2026 - Multi-agent platform for physicians, surgeons, and nurses AI

Project overview

Initum AZ wants a suite of role-aware AI agents to give care teams faster access to protocols, patient context, and follow-up. Focus: safe assistance, traceability, and clear permissions per role.

Role-specific copilots: Agents for physicians, surgeons, and nurses with scoped tools and logging.
Grounded answers: RAG over clinical protocols, OR checklists, device manuals, and care pathways with citations.
Safe and controllable: Guardrails, citations, human confirmation for critical actions, and full audit trail.
Hybrid setup: Azure OpenAI plus on-prem connectors to EHR/LIS/PACS so patient data stays inside.

Initial use cases

AI

Surgical briefing agent

Creates pre-op briefs with risks, key labs, medications, and links to OR checklists.
AI

Nurse shift handover

Summarizes patient status, planned interventions, and alerts so the next shift can act immediately.
AI

Consult preparation for physicians

Generates compact dossiers: timeline, imaging findings, medication changes, and guideline references.
AI

Task routing and alerts

Agent watches signals (lab values, device events) and assigns tasks with clear escalation paths.

Technical approach

AI and orchestration

Azure OpenAI Multi-agent RAG Citations Guardrails

Data and integration

FHIR EHR connector LIS/PACS Event-driven API Gateway

Security and compliance

RBAC Audit logging PII minimization On-prem/hybrid Isolation

Frontend and workplace

Blazor SignalR streaming MAUI Offline-first Role UX

Roadmap 2026

01
Discovery (Jan) - Data mapping, risk assessment, pick pilot use case with OR/ICU.
02
PoC (Feb-Mar) - RAG pipeline on limited sources, one agent live with citations and audit trail.
03
Rollout (Apr-Jun) - Expand roles, task routing, integrate alerts and dashboards.
04
Scale (H2) - Performance, fine-tuning on own data, validation with clinical KPIs.

My role

Hands-on architect operationalizing multi-agent AI safely: from data mapping and guardrails to UX for care teams. Practical, measurable, and trust-focused.

Co-design with clinical experts Citations and audit by default Fast iterations with impact tracking