Inspiration

This project was inspired by the growing intersection of artificial intelligence and healthcare, particularly in preventive diagnostics. With the rapid rise of chronic diseases and the limitations of traditional diagnostic methods, the idea was to build a system that leverages visual biomarkers—such as eye, tongue, and nail images—to provide early health insights.

The goal was to create an accessible and intelligent solution capable of delivering preliminary health assessments, especially in underserved and remote areas. By combining computer vision with real-world telemedicine challenges, the system aims to make early detection more scalable, affordable, and widely available.

What I Learned

Machine Learning & Probabilistic Thinking

Implemented XGBoost for classification along with Bayesian updating to refine predictions dynamically. The probabilistic framework used:

$$ P(D \mid S) = \frac{P(S \mid D)\, P(D)}{P(S)} $$

This formulation updates the posterior probability as new evidence is observed, enabling adaptive confidence scoring and improving diagnostic reliability over time.

Full-Stack Development

Built scalable APIs using FastAPI
Developed a responsive frontend with React + Tailwind CSS
Implemented JWT authentication
Integrated APIs and managed state efficiently

Explainable & Ethical AI

Focused on interpretability using SHAP values, where model predictions can be expressed as:

$$ f(x) = \phi_0 + \sum_{i=1}^{n} \phi_i $$

This helped break down predictions into feature-level contributions, improving transparency and trust in the system. Also explored bias mitigation and privacy-preserving practices for handling sensitive medical data.

Deployment & DevOps

Used Docker for containerization
Managed environments for consistent deployment
Ensured portability of ML-enabled applications

How I Built the Project

The system was designed as a modular full-stack architecture with three key layers:

Backend (FastAPI + SQLAlchemy)

RESTful APIs built using FastAPI
Database management using SQLAlchemy ORM
JWT-based authentication and secure endpoints
Image upload handling and diagnostic processing
Doctor interaction and report generation modules

Frontend (React + Vite + Tailwind)

Modern UI using React with Vite for fast builds
Responsive design with Tailwind CSS
Protected routes and API integration via Axios
Dashboards for users, consultations, and reports

Machine Learning Pipeline

Image feature extraction using OpenCV
Classification using XGBoost
Bayesian updating for probability refinement
SHAP-based explainability integration
End-to-end pipeline from image input to diagnostic output

The entire system was containerized using Docker Compose, ensuring smooth integration and deployment.

Challenges Faced

Database Configuration Issues

Initial PostgreSQL setup led to authentication and connection errors, requiring a shift to SQLite during development and adjustments in schema handling.

Model Performance vs Efficiency

Balancing accuracy with computational efficiency was challenging. XGBoost required careful tuning, and SHAP explanations increased processing overhead.

Frontend–Backend Integration

CORS misconfigurations initially blocked communication between React and FastAPI, resolved through proper middleware setup.

Data Availability & Ethics

Limited access to real medical datasets required the use of synthetic data, highlighting the importance of bias mitigation and real-world validation.

Deployment Complexities

Handling Docker containers, environment variables, and port conflicts required iterative debugging and optimization.

Closing Note

This project demonstrates how AI can move beyond theoretical applications into impactful healthcare solutions. By combining computer vision, probabilistic modeling, and full-stack engineering, it enables accessible and early-stage medical diagnostics while maintaining a strong focus on transparency and ethical AI.