Swasthya Sanket

• 

• 

• 

• 

Inspiration

What it does

India faces a growing burden of Non-Communicable Diseases (NCDs), yet rural areas often lack early detection and personalized care systems. ASHA workers serve as the first line of health intervention but lack tools for real-time risk prediction. Swasthya Sanket aims to bridge this gap using AI-powered mobile and web applications that empower ASHA workers and communities through predictive health assessments, diet/activity recommendations, and early risk alerts.

🩺 What it does Swasthya Sanket is an AI-powered NCD Risk Prediction System with the following features:

Risk Prediction: Predicts risk probabilities of Diabetes, Hypertension, Heart Disease, etc., based on user input.

Personalized Recommendations: Generates a tailored diet chart and daily activity plan.

Voice-enabled Interface: Supports vernacular/Hindi chatbot interaction for ASHA workers.

Real-time Alerts: Sends alerts for high-risk individuals and suggests next steps.

Secure Storage: Uses Firebase for document updates without duplications.

🏗️ How we built it Frontend: Built using Streamlit for offline desktop/mobile deployment; Angular-based web portal under development.

Backend: Python (FastAPI), with ML models trained using NFHS-5 data.

ML Models: Logistic Regression, Random Forest, and XGBoost; converted to ONNX for efficiency.

Database: Firebase Firestore to store and retrieve patient data.

Voice Assistant: Gemini API integrated for Hindi-language chatbot interaction.

Cloud: Integrated with AWS S3 and EC2 for scalable deployment (in progress).

🧱 Challenges we ran into Handling data imbalance and overfitting while training ML models.

Converting scikit-learn models to ONNX and optimizing them for mobile.

Ensuring Firestore document updates are atomic and avoid duplicates.

Building a vernacular-friendly voice chatbot for rural health workers.

Optimizing the app for low-resource devices (limited RAM/storage).

🏆 Accomplishments that we're proud of Built a fully functional offline AI tool that enables ASHA workers to screen villagers instantly.

Achieved 80–85% accuracy in disease risk prediction using real-world datasets.

Integrated a voice-enabled Hindi chatbot for the first time in our projects.

Ensured seamless document updates in Firebase without duplication.

📚 What we learned Advanced model optimization using ONNX.

Firebase Firestore document handling and atomic updates.

Building inclusive UI/UX for non-technical users and health workers.

Real-world ML deployment constraints in rural or low-infra settings.

🔮 What's next for Swasthya Sanket 🌐 Deploy a fully responsive Angular web portal for healthcare officers and administrators.

📊 Add real-time analytics dashboards for NCD spread at village/district level.

🧠 Integrate LLM-based reasoning to suggest treatments or nearest clinics.

📱 Create a Progressive Web App (PWA) for better mobile accessibility.

📡 Enable IoT integration with wearable health devices (e.g., BP monitors).

🏥 Expand to cover maternal health, malnutrition, and infectious diseases.

Built With

• angular-(web-portal-in-progress)-backend:-fastapi-(rest-api)
• built-with-languages:-python
• firebase-hosting-(optional)
• git
• github-(ci/cd)-apis-&-integrations:-google-gemini-api-(for-hindi-vernacular-chatbot)-onnx-(model-export-and-runtime-inference)-tools-&-platforms:-vs-code
• google-colab/jupyter
• html/css-frameworks-&-libraries:-frontend:-streamlit-(offline-desktop/mobile)
• javascript-(typescript-for-angular)
• numpy-databases-&-storage:-firebase-firestore-(for-real-time-document-storage-and-updates)-aws-s3-(for-cloud-based-data-storage)-cloud-services-&-devops:-aws-ec2-(deployment)
• onnx-runtime-(for-optimized-ml-inference)-ml/ds:-scikit-learn
• pandas
• postman
• xgboost