CareBridge AI: Clinical Discharge Orchestration Platform

AI-powered discharge orchestration platform leveraging FHIR R4, MCP, and Google Gemini to generate safe, personalized hospital discharge plans with medication safety verification and care coordination

Comment 2

Project Description (with Markdown + LaTeX support):

CareBridge AI is an intelligent clinical discharge orchestration system designed to enhance patient safety during hospital transitions. The platform addresses a critical gap in healthcare: discharge planning complexity and patient safety risks.

Inspiration & Problem

Hospital discharge is a high-risk transition point where medication errors, missed follow-ups, and poor care coordination lead to preventable readmissions and adverse events. Clinical teams often lack decision support for:

  • Drug interaction detection across medication lists
  • Coordination of specialist follow-ups
  • Patient-friendly discharge instruction generation

Solution Architecture

CareBridge AI integrates three key technologies:

  1. MCP Server (Model Context Protocol) - Provides intelligent context management for clinical reasoning
  2. FastAPI Backend - Exposes FHIR R4 compliant APIs for patient data and discharge workflows
  3. React Frontend - User-friendly dashboard for clinicians to review and approve discharge plans

Technical Highlights

  • FHIR R4 Standard Integration: Fetch patient records from FHIR-compliant servers (tested with HAPI FHIR)
  • Google Gemini 2.0 Flash Integration: Single, unified AI model for:
    • Medication safety analysis
    • Follow-up care gap identification
    • Patient-friendly instruction generation
  • Double-Track Submission: Dual MCP + A2A (Agent-to-Agent) architecture powered by Gemini
  • No External AI Dependencies: Purely Gemini-based reasoning (no Claude, no multi-provider logic)
  • JWT-Protected APIs: Secure clinician authentication and patient data access
  • Demo Patient: Realistic John Martinez case (67M, Type 2 Diabetes, complex medication profile with known Warfarin-Ibuprofen interaction)

Key Features Demonstrated

  1. Load FHIR patient records
  2. Medication safety scanning
  3. Follow-up care review
  4. AI-generated discharge plans
  5. Clinician approval workflow
  6. Admin dashboard for user management

Challenges Overcome

  1. FHIR Complexity - Navigated R4 schema for medication/condition mapping
  2. AI Context Management - Designed MCP protocols for stateful clinical reasoning across service boundaries
  3. Safety-Critical System - Implemented multi-layer verification for medication interactions
  4. React Component Architecture - Built reusable components for real-world healthcare workflows

What We Learned

  • Healthcare interoperability requires careful standard adherence (FHIR R4 conventions)
  • MCP enables flexible AI context management without tight coupling
  • Clinical AI systems demand transparent reasoning and multi-step verification
  • User experience for hospital staff requires minimal cognitive load

Built With

Share this project:

Updates