Technical Deep Dive

FeverDiff AI

AI-powered differential diagnosis assistant for malaria, dengue, chikungunya, and typhoid in Ghana.

The Challenge

"In Ghana, overlapping fever symptoms between malaria, dengue, chikungunya, and typhoid lead to dangerous misdiagnosis. Rural clinics lack specialist support for differential diagnosis."

Project Overview

FeverDiff AI is a clinical decision-support tool designed specifically for Ghana's resource-limited healthcare settings. It uses fine-tuned MedGemma 1.5 with a hybrid AI approach combining LLM reasoning and deterministic clinical rules.

Key Features

MedGemma 1.5 & Gemini Reasoning

Core clinical reasoning powered by specialized medical LLMs for structured diagnosis.

Ghana-Specific Scoring

Tailored logic reflecting regional epidemiological patterns and seasonal disease weighting.

Researcher Workspace

Direct JSON import for rapid test scenario validation and pre-defined clinical test cases.

The Workflow

Patient data entry of 26+ clinical variables

MedGemma handles initial inference

Deterministic Clinical Rule Integration (CRI) validates results

Co-infection alert triggered if probabilities are statistically indistinguishable

Technical Stack

Category	Technologies
AI Models	MedGemma 1.5 4B, Gemini 2.5-Flash
Backend	FastAPI on Modal Labs (GPU T4)
Frontend	Vanilla JS, Glassmorphism CSS

System Architecture

Serverless backend on Modal Labs using GPU T4 for fast inference ( < 2s). Volume caching ensures model weights remain warm.

Project Structure

├── modal_backend.py (Inference Logic)
├── model.py (Clinical Rules)
├── static/ (High-fidelity UI)
└── test_scenarios.json

Development Roadmap

Fine-tune MedGemma 1.5completed

Implement CRI logiccompleted

Deploy to Modal with GPU cachingcompleted

User testing with clinical partnersplanned

Disclaimer

FeverDiff AI is a decision-support tool and NOT a replacement for professional medical diagnosis. Always consult qualified healthcare professionals.