💕 CareLink: AI-Powered Remote Health Monitoring

CareLink is a proof-of-concept project for the AWS Breaking Barriers Hackathon 2025.

It demonstrates how next-generation connectivity (AWS IoT Core), Machine Learning (Amazon SageMaker), and Generative AI (Amazon Bedrock) can combine to deliver real-time, equitable healthcare monitoring — especially for remote and underserved communities.

---

🚀 Solution Architecture

Core AWS Services:

• AWS IoT Core — Ingests vital signs securely via MQTT.
• AWS Lambda — Parses incoming vitals and saves them to storage.
• Amazon DynamoDB — Stores historical patient vitals for retrieval and analysis.
• Amazon SNS — Sends instant alerts if critical thresholds are breached.
• Amazon SageMaker — Predicts the probability of patient instability using machine learning.
• Amazon Bedrock (Titan Text G1 Lite) — Generates clinical-style summaries based on patient history.
• Amazon Bedrock (Nova Sonic) — Have a human like conversation about the data with an Ai Agent.
• AWS CloudWatch — Logs and monitors system operations.

---

🛗 Updated System Overview

1. Vital Collection
   Devices or simulators publish patient vitals (heart rate, blood oxygen, temperature) to the MQTT topic carelink/vitals.

2. AWS IoT Core ➔ Lambda
   An IoT rule triggers the CareLinkVitalsProcessor Lambda on every new vital sign message:

   • Parses the data.
   • Saves the raw vitals directly to DynamoDB.
   • If critical thresholds are breached, sends an SNS alert immediately.

3. Frontend Dashboard
   The React dashboard fetches historical vitals from DynamoDB by calling a separate Lambda:

   • Sends the latest 24 hours of vitals to SageMaker for a stability prediction.
   • Sends the latest 24 hours of vitals to Bedrock for an AI-written clinical summary.
   • Displays:
     • Instability Risk Meter (with live percentage)
     • Raw prediction probability (exact model output)
     • Bedrock AI Clinical Summary
     • Graphs of vitals over time for the past 3 months of data
     • Have a human like conversation about the data with an Ai Agent

---

🧠 Machine Learning Model: SageMaker Stability Classifier

Training Overview:

• We created a clinically realistic synthetic dataset using engineered thresholds.
• Features:
  • Heart Rate (normalized)
  • Blood Oxygen (normalized)
  • Temperature (normalized)
• Output:
  • Binary label: Stable (0) vs Unstable (1)

Focus:
⚡ Emphasize safety by minimizing false negatives (missing real instability is worse than sending a false alert).

Model Details:

• Type: Binary classification
• Algorithm: XGBoost
• Training Location: SageMaker in us-east-1

---

📈 Feature Table

Feature Name	Type	Description	Range
heart_rate_normalized	Continuous	Normalized patient heart rate	0.0 – 1.0
blood_oxygen_normalized	Continuous	Normalized blood oxygen saturation	0.0 – 1.0
temperature_normalized	Continuous	Normalized body temperature	0.0 – 1.0
label	Target	1 = unstable patient, 0 = stable patient	0 or 1

---

🗄️ DynamoDB Storage

Table Name: carelink_alerts
Schema:

Field	Type	Description
device_id	String	Patient's device ID
timestamp	String (ISO 8601)	Exact time of the reading
heart_rate	Number	Heart rate (bpm)
blood_oxygen	Number	Blood oxygen (%)
temperature	Number	Temperature (°C)
status	String	"stable" or "unstable" label for basic flagging

✅ Only raw vitals + status are stored — no SageMaker predictions or Bedrock summaries saved.

---

📋 Updated System Diagram

  Device/Simulator
       |
  [ MQTT Publish ]
       |
       v
  AWS IoT Core (carelink/vitals)
       |
  IoT Rule
       |
       v
  AWS Lambda (CareLinkVitalsProcessor)
  - Save raw vitals to DynamoDB
  - Send SNS alerts if critical
       |
       v
  AWS DynamoDB (carelink_alerts)

  (Separately)
  
  Frontend Dashboard
       |
  ➔ Lambda (Fetch vitals)
       |
  ➔ SageMaker (Predict risk based on latest 24h vitals)
       |
  ➔ Bedrock (Summarize last 3 months of vitals)
       |
  ➔ Bedrock (Nova Sonic) (Have a human like conversation about the data with an Ai Agent)
       |
  ➔ React Frontend (Display meter, graph, raw score, AI summary)

---

🧠 Bedrock AI Clinical Summaries

• Model: Titan Text G1 Lite
• Prompt Strategy:
  • Summarize vitals history factually.
  • Highlight increases, decreases, and trends.
  • Avoid guessing or proposing clinical treatments.
  • Output designed for quick review by healthcare workers.

---

📋 Frontend Features (React)

Feature	Details
Instability Risk Meter	Animated circular meter showing SageMaker prediction
Raw Probability Display	Exact model output shown alongside the meter
Bedrock AI Summary	Auto-generated clinical summary of patient's history
Vitals Over Time Graph	Smooth, clean Chart.js graphs (no background fills)
Developer Input Panel	Bulk upload of test JSON vitals (accordion panel)

---

📚 Why Normalize Vitals?

• Different clinical signs (HR, SpO2, Temp) operate at different numeric scales.
• Normalization ensures fair weighting during model training.
• Prevents HR from overwhelming oxygen and temperature in model influence.

---

🌟 Future Enhancements

• Real wearable integration (BLE/5G)
• Admin dashboard for multi-patient management
• Mobile app with SNS Push
• Fine-tuned Bedrock prompts for different clinical personas (nurse vs doctor)

---

🏁 Final Thoughts

CareLink shows how AI + IoT + Cloud can bridge the healthcare access gap —
bringing trusted, automated support to frontline healthcare workers everywhere.

---

📜 License

MIT License — fork, build, and improve freely!

---

🔗 Important Links

• Demo Video
• Demo Script
• Try the app, follow the README steps to start Nova Sonic, then, the app will also open wiht Nova integrated

---

✍️ Author

• GitHub
• LinkedIn
• Devpost

---

🚀 CareLink — Democratizing Remote Healthcare Through AI + IoT

---