Grameen AI Health Triage

A Databricks-powered AI platform for symptom triage and medicine demand forecasting.

Comment

🏥 Grameen AI Health Triage + Medicine Demand Predictor

🌟 Inspiration

In many rural parts of India, access to healthcare is still a major challenge. Patients often delay treatment due to lack of nearby doctors, and primary health centers frequently face medicine shortages.

We wanted to build a solution that could assist patients in early decision-making while also helping healthcare systems prepare in advance.

This project was inspired by the idea that AI can bridge both gaps — diagnosis and availability — simultaneously.

💡 What It Does

Our system provides a dual-layer AI solution:

🧑‍⚕️ 1. Health Triage System

  • Users input their symptoms (e.g., fever, headache, vomiting)

  • The AI model classifies severity into:

    • Low
    • Medium
    • High

  • Supports Indic inputs like "bukhar sar dard"

💊 2. Medicine Demand Predictor

  • Uses historical demand data
  • Predicts future medicine requirements
  • Helps prevent stockouts in rural health centers

🛠️ How We Built It

🔹 Data & Processing (Databricks)

  • Used Apache Spark (PySpark) for distributed data processing
  • Stored structured data using Delta Tables (Lakehouse architecture)
  • Managed data pipelines directly inside Databricks

🔹 AI / ML Pipeline

We built an NLP-based classification model:

  • Tokenization: $$ text \rightarrow words $$
  • Feature extraction using HashingTF
  • Label encoding using StringIndexer
  • Classification using Logistic Regression

Final pipeline:

Symptoms → Tokenizer → HashingTF → Logistic Regression → Severity Prediction

🔹 Medicine Prediction Logic

  • Aggregated historical demand data
  • Used statistical averaging to estimate future needs: $$ Predicted\ Demand = \frac{\sum Demand}{n} $$

🔹 Frontend & API

  • Built a simple UI for user interaction
  • Backend API handles prediction requests
  • Modular architecture:
Frontend → API → Databricks Model → Response

🚧 Challenges We Faced

  • Limited dataset → We simulated realistic data to demonstrate scalability

  • Label mismatch issue → Fixed by dynamically mapping labels using StringIndexer

  • Databricks serverless restrictions → Switched from DBFS paths to managed Delta tables

  • Balancing speed vs accuracy → Focused on a working prototype rather than perfect model accuracy

📚 What We Learned

  • How to build end-to-end AI pipelines using Databricks
  • Importance of data preprocessing and feature engineering
  • Real-world architecture: separating data layer, API layer, and UI
  • Handling practical issues like label encoding and deployment constraints

🌍 Impact

This system can:

  • Reduce unnecessary hospital visits
  • Enable early risk detection
  • Ensure better medicine availability in rural areas

🚀 Future Scope

  • Integration with real healthcare datasets
  • Voice-based input for accessibility
  • Deployment in government health systems
  • Real-time model serving using APIs

🏁 Conclusion

Grameen AI Health Triage is a step toward accessible, scalable, and intelligent healthcare for rural India, combining AI-driven diagnosis with proactive resource planning.

Built With

Share this project:

Updates