README.md

metadata

title: SheGuard
emoji: 🤱
colorFrom: pink
colorTo: purple
sdk: docker
pinned: false
short_description: Maternal mortality early warning using Mamba3 SSM AI

SheGuard — Maternal Risk Assessment

AI-powered maternal mortality early warning system, built with Mamba3 Sequential State-Space Models and WHO clinical safety rules.

Features

• Mamba3 SSM model — Temporal sequence analysis of prenatal vital signs
• 3-tier alert system — GREEN / AMBER / RED risk classification
• WHO clinical safety net — Hard rule overrides for obvious danger signs
• OCR auto-fill — Photograph a paper prenatal record card to auto-populate visit data
• Resource-aware routing — Generates transfer orders when clinic lacks blood supply or staff
• Explainable AI — Shows top contributing features for each prediction

How to Use

1. Enter patient vitals from prenatal visits (or upload a photo of the record card)
2. Click "Assess maternal risk"
3. View the risk level, contributing factors, and recommended clinical actions

Architecture

Patient vitals → Mamba3 SSM (5-visit sequence) → Risk prediction
                                                    ↓
                                WHO Clinical Safety Rules
                                                    ↓
                                    GREEN / AMBER / RED alert

Tech Stack

• Model: PyTorch, Mamba3 SSM (Trapezoidal discretization + Complex state + MIMO)
• API: FastAPI + Uvicorn
• Frontend: HTML/CSS/JS dashboard
• OCR: OpenCV + Pytesseract
• Dataset: UCI Maternal Health Risk (1,014 samples)