RuralShield

A public health intelligence platform for maternal emergencies + population surveillance

Comment

About RuralShield

Inspiration

Every 2 minutes, a woman dies from preventable pregnancy-related causes, and nearly 90% of infectious disease deaths occur in low-resource settings.

What stood out to us was that these deaths are not due to lack of medical knowledge — but due to system failures, especially the Three Delays:

  1. Delay in seeking care
  2. Delay in reaching the right facility
  3. Delay in receiving treatment

We were inspired to build a solution that directly supports frontline health workers, who often operate with limited tools, no real-time data, and poor connectivity.

What We Built

RuralShield is a dual-tier public health intelligence system designed specifically for low-resource environments.

It combines:

  • 🧠 AI-powered maternal risk prediction
  • 🚑 Facility capability matching
  • 🌍 Population-level surveillance dashboards
  • 📱 SMS-based accessibility for offline regions

⚙️ How We Built It

1. Machine Learning (Maternal Risk Prediction)

We trained an XGBoost classifier on maternal health data with features such as:

  • Age
  • Blood Pressure (Systolic & Diastolic)
  • Blood Sugar (BS)
  • Body Temperature
  • Heart Rate

The model predicts risk levels:

  • Low Risk (0)
  • Mid Risk (1)
  • High Risk (2)

We optimized the model using hyperparameter tuning

2. Explainable Action Engine

Instead of just predicting risk, we built a rule + ML hybrid layer that:

  • Identifies key contributing factors
  • Generates personalized clinical recommendations
  • Produces a referral slip instantly

3. Facility Matching System

We created a dataset of healthcare facilities with capabilities like:

  • ICU, NICU, Blood Bank, C-section

The system:

  • Filters facilities based on required services
  • Estimates travel time
  • Prevents misdirected referrals

4. Public Health Dashboard

We designed an interactive dashboard using Streamlit to:

  • Detect malnutrition, anemia, and food insecurity
  • Identify high-risk rural regions
  • Track infectious diseases (Malaria, Dengue, TB, COVID)
  • Simulate transmission networks using graph models

5. Low-Resource Design

To ensure real-world usability:

  • The system runs locally (no cloud dependency)
  • Designed to support SMS-based triage
  • Works in areas with limited or no internet connectivity

What We Learned

  • How to integrate machine learning with real-world healthcare workflows
  • The importance of Explainable AI (XAI) in clinical decision-making
  • Designing for low-resource environments requires simplicity, not complexity
  • Combining individual-level prediction with population-level insights creates stronger impact

Challenges We Faced

1. 🔌 Low-Resource Constraints

Designing a system that works without internet forced us to rethink:

  • Deployment strategies
  • Data access
  • User interaction

2. Balancing ML and Interpretability

We needed predictions to be:

  • Accurate and
  • Understandable for health workers

This led us to combine ML with rule-based explanations.

3. Data Limitations

  • Limited real-world datasets for rural health scenarios
  • Needed to simulate realistic conditions for testing

4. Time Constraints (Hackathon)

Building an end-to-end system (ML + dashboard + matching + UX) within limited time required:

  • Prioritization of high-impact features
  • Rapid prototyping

Impact

RuralShield transforms fragmented healthcare into a connected decision system:

  • Faster triage decisions
  • Smarter referrals
  • Early outbreak detection
  • Data-driven policy planning

🌱 Why It Matters

This project is not just about technology —
it’s about reducing inequality in healthcare access.

By empowering frontline workers and improving system-level visibility,
RuralShield helps ensure that no life is lost due to delays in care.

Share this project:

Updates