Severity Ranker 🏥

An AI triage system that ranks incoming emergency patient cases by severity — combining a Logistic Regression classifier for structured scoring with LLM-based reasoning (OpenAI) for dynamic, edge-case handling, in a simulated hospital environment.

What it does

• Severity classification — a trained Logistic Regression model scores patient cases from structured features.
• LLM edge-case reasoning — for ambiguous or borderline cases, an LLM adds context-aware judgment beyond the classifier's fixed decision boundary.
• Batch intake — upload a CSV of patient records; the app processes and ranks them.
• Web interface — Flask app with user login for reviewing assessments.

How it works

1. Patient features are preprocessed (pandas) and scaled with a persisted scaler.pkl.
2. The Logistic Regression model (model.pkl) produces a severity score/class.
3. Borderline cases are escalated to the LLM for a reasoned second opinion.
4. Results are surfaced in the UI, ranked by severity.

Tech stack

Python · Flask · scikit-learn (Logistic Regression) · pandas · joblib · OpenAI API · SQLite

Getting started

git clone https://github.com/baghdme/severityranker
cd severityranker

python -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install flask pandas scikit-learn joblib openai python-dotenv

cp .env.example .env        # then add your OPENAI_API_KEY
python app.py

Configuration

Set these in a .env file (never commit it — it's gitignored):

Variable	Purpose
OPENAI_API_KEY	LLM reasoning for edge cases
FLASK_SECRET_KEY	Flask session signing

Notes

• Built for the EECE 490 (Machine Learning) course project at AUB.
• Data is simulated — no real patient information.
• Demo login users are seeded for local testing only; change them before any real deployment.

---

Built by Mohamad Baghdadi.