Provana Acutecare Copilot

Emergency clinicians make life-or-death decisions in minutes, often across fragmented systems with no unified operational workflow. Provana AcuteCare Copilot was built to address that problem, an agentic emergency-care assistant that transforms a plain-language patient description into a live, interactive clinical workflow rendered directly into the interface as the AI responds.

Built on CopilotKit’s AG-UI framework, the system goes beyond traditional chat interfaces. When a clinician enters a prompt like “56-year-old febrile hypotensive patient with penicillin allergy,” the AI does not return static text. Instead, it calls a renderProtocol action that streams a fully interactive protocol card into the UI in real time. Vitals populate progressively, severity states update dynamically, countdown timers activate for bundle tasks, and treatment plans adapt automatically based on patient context.

One of the core features is the allergy-aware treatment mutation engine. When a penicillin allergy is detected, the standard antibiotic recommendation is automatically crossed out and replaced with a safer alternative, including dosage guidance and safety rationale, before any action is taken by the clinician.

Every action is governed through role-based operational controls. Nurses can triage and initiate workflows, doctors can authorize medications, and only attending physicians can approve high-risk interventions like tPA. If a restricted action is attempted, the governance layer intercepts it in real time and streams BLOCKED, ROUTED, or APPROVED events into a live control panel with full audit visibility.

The system also supports agentic follow-up behavior. After an authorized action is completed, the AI automatically asks the next clinically relevant question without requiring the user to re-prompt, helping maintain clinical workflow continuity.

Supported protocols include:

• Sepsis Bundle (SEP-1)
• Stroke Code
• Pediatric Fever

Tech stack:

• Next.js 16 App Router
• CopilotKit 1.57 (useCopilotAction, useCopilotReadable, useCopilotChat, CopilotRuntime)
• OpenAI GPT-4o
• Tailwind CSS
• Zustand

Provana AcuteCare Copilot demonstrates how generative UI can evolve beyond conversational AI into adaptive operational interfaces for high-acuity clinical environments.

This project was built from scratch during the hackathon, with no prior code reused. Key components developed include the protocol renderers, the allergy mutation engine, the governance layer, the bundle checklist, and action buttons. The project was initially set up using create-next-app and the standard CopilotKit configuration.