CardioCommand

AI-powered post-cardiac surgery monitoring and care coordination platform.

Built for the Healthcare AI Hackathon. Solo build. Demo-first architecture.

---

What It Does

CardioCommand closes the communication gap between hospital discharge and home recovery for cardiac surgery patients.

MD Dashboard — Dark, clinical, data-dense. For cardiologists and care coordinators.

• 12 live vitals metrics streamed via WebSocket at 10 readings/sec
• LangGraph AI agent: RAG → Vitals Analysis → Risk Scoring → Action Generation
• Ambient visit documentation: live transcript → SOAP note via GPT-4o
• War Room: population risk overview with sortable patient table
• Demo scenario control panel

Patient App — Warm, mobile-first, conversational. For the patient at home.

• Live vitals in plain English (not medical jargon)
• Cora AI companion: GPT-4o chat with escalation detection
• Medication tracker, recovery progress bar, plain-English recovery plan
• Syncs alerts in real-time to MD Dashboard

---

Tech Stack

Layer	Technology
Frontend	React 18 + Vite + Tailwind CSS + Framer Motion + Recharts
Backend	Python FastAPI + LangGraph + OpenAI GPT-4o
Real-time	WebSockets (FastAPI native)
AI/ML	GPT-4o, FAISS RAG, Deterministic risk model
Deployment	Vercel (frontends) + Railway (backend)

---

Quickstart

1. Backend

cd backend
pip install -r requirements.txt

# Add your OpenAI API key
echo "OPENAI_API_KEY=sk-your-key-here" > .env

# Build the RAG index (run once)
python -c "from rag.indexer import build_index; build_index()"

# Start the server
uvicorn main:app --reload --port 8000

2. MD Dashboard

cd apps/md-dashboard
npm install
npm run dev
# → http://localhost:5173?demo=true

3. Patient App

cd apps/patient-app
npm install
npm run dev
# → http://localhost:5174?demo=true

---

Demo Setup (8-Minute Presentation)

1. Open http://localhost:5173?demo=true in Tab 1 (MD Dashboard)
2. Open http://localhost:5174?demo=true in Tab 2 (Patient App)
3. Open patient app on phone at the same URL
4. Both apps default to John Mercer — Day 8 Early Warning scenario

Demo Flow

• 0:00 — The hook: "1 in 5 cardiac patients readmitted within 30 days..."
• 0:45 — Patient app on phone → Chat with Cora → report chest soreness
• 2:15 — MD Dashboard alert fires in real-time
• 3:15 — Click "Run Full Analysis" → watch LangGraph pipeline stream
• 5:15 — Ambient Documentation → speak patient conversation → SOAP note builds live
• 7:15 — War Room → $98K predicted savings
• 7:45 — Close

---

API Endpoints

GET  /patients                     → All patients with live vitals
GET  /patients/{id}                → Full patient record
GET  /patients/{id}/timeline       → Event log
WS   /vitals/stream/{patient_id}   → Live vitals (10 readings/sec)
GET  /vitals/{id}/history          → Last N readings
POST /ai/analyze                   → LangGraph agent (SSE)
POST /ai/chat                      → Cora patient chat (SSE)
POST /ai/pre-visit-brief           → Pre-visit intelligence brief (SSE)
POST /ai/soap-note                 → SOAP note from transcript
GET  /demo/scenarios               → List scenarios
POST /demo/set-scenario            → Switch scenario
POST /demo/trigger-alert           → Fire an alert
WS   /demo/events                  → Demo event broadcast

---

Architecture

Patient Wearable (simulated)
         ↓ WebSocket (10hz)
   FastAPI Backend
         ├── Vitals Simulator (scenario-driven, Gaussian noise)
         ├── LangGraph Agent
         │   ├── [retrieve_guidelines] — FAISS RAG
         │   ├── [analyze_vitals] — GPT-4o
         │   ├── [score_risk] — Rule-based model
         │   ├── [decide_alert] — Conditional router
         │   ├── [generate_outreach_script] — Medium risk
         │   ├── [generate_urgent_brief] — High/Critical risk
         │   └── [generate_summary] — Final output
         └── Demo Control (WebSocket broadcast)
              ↓                     ↓
      MD Dashboard           Patient App
   (localhost:5173)        (localhost:5174)

---

Mock Patients

Patient	Surgery	Day	EF	Scenario
John Mercer, 67	CABG	8	35%	Early Warning (Risk 71)
Rosa Delgado, 54	TAVR	5	52%	AFib Detected (Risk 94)
Marcus Webb, 71	ICD	13	28%	Pre-Visit (Risk 55)
Sarah Kim, 48	Mitral Valve Repair	30	62%	Full Recovery (Risk 14)

---

CardioCommand — Healthcare AI Hackathon 2025