SwasthAI Guardian

Live Sanitary Pad Request (Alerts NGO Workers)
Villager Home Dashboard
Secure Role-Based Login
Live GPS Ambulance Request
Services Introduction Page
NGO Dashboard
Hospital Admin Dashboard

Inspiration

Over 65% of India’s population resides in rural areas, yet access to healthcare infrastructure remains severely limited. We were inspired by ASHA (Accredited Social Health Activist) workers who tirelessly serve these communities, often relying on manual records and limited clinical tools.

Our goal was to build a scalable, AI-powered digital health ecosystem that bridges this gap providing villagers with immediate, accessible medical guidance in their native language, while equipping frontline workers and district authorities with real-time data for proactive healthcare decisions.

What it does

SwasthAI Guardian is an offline-first, role-based healthcare platform connecting Villagers, ASHA/NGO workers, and Hospital Administrators into a unified system.

For Villagers:

Multilingual, voice-enabled AI Symptom Checker powered by a ML model (Random Forest) trained on rural disease patterns (e.g., Malaria, Dengue, Cholera)
"Sakhi" – a private conversational AI assistant for women's health support
AI Skin Disease Scanner – on-device image analysis for preliminary dermatological screening
Emergency Request System – enables quick assistance requests even in low-connectivity areas

For ASHA/NGO Workers:

Dashboard to monitor high-risk pregnancies using WHO-based scoring
Child malnutrition tracking
Management of emergency and sanitary requests

For Hospital Admins:

District-level analytics dashboard
Detection of 24-hour symptom clusters for early outbreak alerts
Downloadable reports for operational insights

How we built it

We designed SwasthAI Guardian as a modular, resilient three-tier architecture:

Frontend:

React 18 + Vite
Tailwind CSS with custom design system
Offline-first support using Service Workers and IndexedDB

Backend:

Node.js + Express
SQLite for lightweight deployment
JWT authentication with bcryptjs password hashing

AI Service:

Python FastAPI microservice
Custom-trained Random Forest Classifier (~91.3% cross-validated accuracy)
Groq-powered LLM for conversational AI ("Sakhi")

Challenges we ran into

Implementing a truly offline-first system was our biggest challenge. We had to:

Store interactions locally without internet
Sync data automatically when connectivity is restored
Maintain consistent system state across offline and online modes

Additionally, optimizing the ML model for both accuracy and performance required multiple iterations to ensure real-time usability.

Accomplishments that we're proud of

Built a custom ML diagnostic model with ~91.3% accuracy tailored to rural healthcare
Developed a fully localized interface supporting English, Hindi, Marathi, Tamil, and Bengali
Created a working multi-role healthcare workflow (User → NGO → Admin)
Delivered a functional offline-capable system for real-world rural use