DigitalTwinAI

An AI-powered digital health twin that turns blood reports into WHO-aligned, explainable risk insights before disease is diagnosed.

Comment

DigitalTwinAI

Explainable AI for Preventive Healthcare

Inspiration

India is facing a silent but escalating health crisis — not because treatment is unavailable, but because chronic diseases are detected far too late.

  • Heart disease accounts for over 28% of all deaths
  • More than 57% of diabetes cases remain undiagnosed
  • Nearly 60% of thyroid disorders go unnoticed

In most cases, people only realize something is wrong after complications begin.

During our research, we observed a recurring pattern: people do get blood tests done, but they struggle to understand what those numbers actually mean. Lab reports are fragmented, highly technical, and lack context. Doctors, constrained by time, often don’t have access to a complete longitudinal health picture either.

This gap between data availability and data understanding inspired us to build DigitalTwinAI — a system focused on early awareness, explainability, and prevention, grounded in trusted medical standards.

What It Does

DigitalTwinAI is a WHO-aligned AI Health Digital Twin that helps users interpret blood reports and understand long-term health risks in a clear, explainable, and non-diagnostic manner.

The platform enables users to:

  • Upload real blood reports (PDF or image)
  • Extract key clinical values using OCR and AI-assisted parsing
  • Validate results against WHO South Asia clinical guidelines
  • View quantified risk estimates for:
    • Diabetes
    • Hypertension
    • Cardiovascular disease
  • Ask follow-up questions through a constrained health assistant
  • Download a physician-ready summary for awareness and discussion

The system is designed strictly for risk awareness and prevention, not diagnosis.

How We Built It

DigitalTwinAI was developed as a modular, end-to-end system:

  • Frontend: React (Vite + Tailwind) for a clean, focused UI
  • Backend: FastAPI (Python) to orchestrate ingestion, validation, prediction, and explanation
  • OCR: Tesseract for extracting text from real-world lab reports
  • WHO Validation Engine: Rule-based logic using WHO South Asia clinical thresholds stored in JSON
  • Machine Learning: Random Forest models trained offline to estimate disease risk using interacting clinical and lifestyle features
  • AI Layer (Gemini 1.5 Flash): Used only for:
    • Structured data extraction from OCR output
    • Human-readable explanations of risk results
    • Generating physician-ready summaries
  • Deployment: Dockerized frontend and backend for reproducible execution

All medical risk calculations are handled by deterministic logic and trained ML models. The AI does not make medical decisions.

Challenges We Faced

  • Parsing inconsistent and noisy real-world lab report formats
  • Aligning ML predictions with WHO classifications without overstating or understating risk
  • Ensuring AI explanations were helpful yet clearly non-diagnostic
  • Reliable frontend–backend integration under tight time constraints
  • Designing for explainability instead of black-box outputs

Accomplishments

  • Built a fully functional end-to-end system within a 24-hour hackathon
  • Successfully integrated OCR, WHO rules, ML prediction, and AI explanation into a single workflow
  • Prioritized trust, transparency, and interpretability over raw prediction accuracy
  • Generated physician-ready summaries users can realistically share
  • Deployed the entire stack using Docker for clean, reproducible execution

What We Learned

  • Execution, clarity, and communication matter as much as the idea itself
  • In healthcare, explainability is essential for trust
  • AI is most powerful when it supports understanding, not replaces judgment
  • Building responsibly means knowing where not to use AI

What’s Next for DigitalTwinAI

  • Expanding the digital twin to support longitudinal health tracking
  • Integrating wearables and continuous vital signals
  • Adding vernacular language support for broader accessibility
  • Conducting clinical validation studies with partner clinics
  • Exploring secure integration with national digital health ecosystems

DigitalTwinAI is a step toward making preventive healthcare more understandable, accessible, and trustworthy — especially in settings where early awareness can significantly change outcomes.

Built With

Share this project:

Updates