HealthRisk Atlas

Clinical signal workspace

Project context

A health screening interface built to read like a briefing, not a generic dashboard.

The product combines structured intake, predictive scoring, result review, and longitudinal history in one frontend connected to a FastAPI service and machine learning pipeline.

Workflow

End to end

Assessment, scoring, review, history, and analytics share one UI system.

Model shape

1 + 3

Three disease streams feed one composite signal.

Clinical stance

Supportive

Useful for early interpretation, never a replacement for diagnosis.

Methodology

Training sources and modeling strategy

The frontend is explicit about what informs the predictions so users can judge credibility and fit.

PIMA Indians Diabetes Database
Cleveland Heart Disease Dataset
Framingham Heart Study
Stroke Prediction Dataset

Each disease stream can include multiple candidate models and ensemble logic, with explainability and recommendations layered on top of the prediction workflow.

Technology stack

Current build architecture

The redesign focuses on frontend cohesion, but the interface remains aligned with the actual backend and model workflow.

Frontend

  • Next.js 16.1.6
  • React 19.2.4
  • TypeScript 5.9
  • Tailwind CSS

Backend

  • FastAPI
  • SQLAlchemy
  • SQLite or PostgreSQL-backed storage
  • scikit-learn model pipeline

Important disclaimer

This is not medical advice

The application is designed for education, exploration, and clinical decision support workflows. It does not diagnose disease and should not be used as a substitute for professional care.

Model limitations

Where caution matters most

Predictions are statistical estimates and may not reflect every clinical nuance.
Model performance depends on the quality and recency of the data entered.
Population bias can affect generalization across demographics or care settings.
Family history and structured inputs do not capture the full genetic or environmental picture.