---

Emergency AI Healthcare Companion

An AI-powered, multi-agent emergency response system that detects medical anomalies in real time and delivers reliable alerts across multiple channels, designed with real-world healthcare constraints in mind.

---

🩺 Problem Statement

Medical emergencies such as severe tachycardia, hypoxia, or sudden vitals deterioration often go unnoticed until it is too late. Existing systems are typically:

• Reactive instead of proactive
• Dependent on a single notification channel
• Fragile under real-world constraints (telecom regulations, network failures, regional restrictions)

Goal: Build a reliable, intelligent, and explainable emergency-response companion that can detect anomalies early and ensure alerts reach caregivers without a single point of failure.

---

Our Solution

We propose an AI-driven, multi-agent healthcare companion that:

1. Continuously monitors patient vitals (simulated / real-time)
2. Detects anomalies using intelligent agents
3. Performs medical severity classification (triage)
4. Identifies nearby hospitals and estimates ETA
5. Sends alerts via redundant notification channels
6. Gracefully handles real-world failures (SMS blocking, auth issues, missing channels)

Reliability is treated as a first-class design constraint, not an afterthought.

---

System Architecture (High-Level)

Vitals Stream
     ↓
Vital Monitor Agent
     ↓
Medical Agent (Risk + Triage)
     ↓
Orchestrator Agent
     ↓
Routing Agent ──→ Nearest Hospital + ETA
     ↓
Notification Agent
     ↓
SMS | WhatsApp | Push | Webhook

Each agent is loosely coupled, independently testable, and replaceable.

---

Agents Overview

1. Vital Monitor Agent

• Continuously processes vitals (HR, SpO₂, etc.)
• Detects deviations from patient baseline

2. Medical Agent

• Performs medical triage and severity classification
• Generates explainable medical reasoning using an LLM

3. Orchestrator Agent

• Central decision-maker
• Coordinates agents and ensures fail-safe execution

4. Routing Agent

• Identifies nearest medical facilities
• Estimates ETA based on location

5. Notification Agent

• Sends alerts across multiple channels
• Implements graceful fallback logic

---

Notification Strategy (Real-World Aware)

Supported Channels

• SMS (Twilio – demo only)
• WhatsApp (Twilio – demo only)
• Push Notifications (planned)
• Webhooks (supported)

Why Multi-Channel Matters

In many regions (e.g. India), SMS delivery is regulated (DLT) and may fail silently.

Our system:

• Differentiates “SMS sent” vs “SMS delivered”
• Treats SMS as best-effort, not guaranteed
• Is designed to fall back to other channels automatically

This reflects real-world deployment readiness, not idealized assumptions.

---

Important Note on Twilio (Demo-Only)

Twilio is used temporarily for demonstration and testing purposes only.

• SMS / WhatsApp delivery may be restricted by regional regulations

• Phone numbers must be verified on Twilio (trial accounts)

• In real-world deployments, this module would be replaced with:

  • Government-compliant SMS gateways
  • Hospital paging systems
  • Secure push notification services

The system architecture is provider-agnostic — Twilio is not a hard dependency.

---

Tech Stack

Backend

• Python
• FastAPI
• Uvicorn
• SQLite (local auth DB)
• Twilio API (temporary demo integration)
• Ollama (local LLM inference)

Frontend

• React
• Vite
• Modular component-based architecture

AI / LLM

• Ollama (local inference)

• Used for:

  • Medical reasoning
  • Triage explanation
  • Decision justification

---

Demonstration Flow

1. Vitals cross a critical threshold

2. System detects anomaly

3. Medical agent classifies severity (e.g. Severe Tachycardia – High Risk)

4. Routing agent identifies nearest hospital + ETA

5. Alerts are dispatched via:

   • Webhooks
   • SMS / WhatsApp (best-effort, demo only)

6. All actions are logged and visible on the dashboard

---

Environment Configuration

Create a .env file in the project root:

TWILIO_SID=ACxxxxxxxxxxxxxxxxxxxx
TWILIO_TOKEN=xxxxxxxxxxxxxxxxxxxx
TWILIO_NUMBER=+1xxxxxxxxxx
EMERGENCY_PHONE=+91xxxxxxxxxx

🔹 Notes

• .env is not committed to the repository
• Twilio credentials are optional if SMS is not required
• The system will continue functioning even if SMS is disabled

---

How to Run the Project

1 Start Ollama (LLM Server)

Ollama must be running in the background:

ollama serve

Ensure your required model is available:

ollama pull llama3

llama run phi3

and then type hello

2️ Start Backend Server

From the project root:

pip install -r requirements.txt
uvicorn backend.main:app --reload --port 8000

---

3️ Start Frontend Server

cd frontend
npm install
npm run dev

Frontend runs at:

http://localhost:5173

---

Services That Must Be Running

Service	Purpose
Ollama	LLM inference for medical reasoning
FastAPI backend	Core logic & agents
Frontend (Vite)	Dashboard & visualization
Twilio (optional)	Demo SMS / WhatsApp

---

Key Innovations

• Multi-agent intelligence (not a monolith)
• Reliability-first system design
• Explainable medical reasoning
• Graceful degradation under failures
• Designed for real-world healthcare constraints

---

Team

Nihar Sagar G
Seepana Rishita 

Emergency AI Healthcare Companion

• Interdisciplinary collaboration across AI, systems, and frontend
• Built with scalability, ethics, and reliability in mind

---

Final Note to Judges

Healthcare systems must function when conditions are worst.

This project is not just a prototype — it is a resilient, thoughtful blueprint for intelligent emergency response systems.

Thank you for your time and consideration.

---