Medical emergency information at a tap.

• NFC bracelet + hub: Patient wears a bracelet with a unique 15-character User ID. When tapped on the NFC scanner hub, the hub sends the ID over serial to the dashboard.
• Doctor portal (web): Clinics register at-risk patients and assign NFC bracelets. Patient data is stored encrypted; only the User ID lives on the tag (no PHI on the bracelet).
• EMT dashboard: Staff connect the NFC hub to their computer, open the dashboard, and tap a bracelet. They see patient info immediately, plus an AI summary and a risk/seriousness score from a trained model, all without on-the-spot rediagnosis.

Web app layout: webapp/config/ (Django settings), webapp/accounts/ (doctor auth), webapp/nfc_users/ (patients, NFC mapping, API, AI overview), webapp/risk_scoring/ (features, training, service), webapp/frontend/ (Next.js app).

• Python 3.11+
• Node.js 20+ and npm
• (Optional) NFC hub connected via USB for dashboard tap-to-read

From the repo root:

cd webapp
python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

pip install -U pip
pip install -r requirements.txt

Environment: Create webapp/.env:

AI_OVERVIEW_API_KEY=your_ark_api_key
AI_OVERVIEW_BASE_URL=https://api.ark-labs.cloud/api/v1
AI_OVERVIEW_MODEL=your_fast_model_name

(Do not wrap values in quotes. Optional if you only need risk scores.)

Database:

python manage.py migrate

Create a doctor user (optional):

python manage.py create_doctor doctor@example.com yourpassword --first-name "Doctor" --last-name "User"

The seriousness/risk score uses a scikit-learn model trained on the UCI diabetic dataset. Run once (or after pulling new code/data):

# From webapp, with venv activated
python manage.py train_risk_model --min-rows 25 --min-positives 5 --max-rows 50000

• Training data: webapp/risk_scoring/data/diabetic_data.csv
• Artifacts: webapp/risk_scoring/artifacts/risk_model_risk-v3-*.joblib

Custom CSV:

python manage.py train_risk_model --csv-path /path/to/data.csv --max-rows 10000

Check which model is loaded:

python manage.py shell -c "from risk_scoring.service import RiskScoringService; print(RiskScoringService()._load_latest_model_payload().get('model_version'))"

From the repo root:

cd webapp/frontend
npm install

Create webapp/frontend/.env.local if needed:

NEXT_PUBLIC_DJANGO_API_BASE_URL=http://127.0.0.1:8000

Terminal 1 – backend:

cd webapp
.venv\Scripts\activate   # Windows
# source .venv/bin/activate   # macOS/Linux
python manage.py runserver

Terminal 2 – frontend:

cd webapp/frontend
npm run dev

• API: http://127.0.0.1:8000
• App: http://localhost:3000 (or the port Next.js shows)

• Hardware: Arduino Nano, PN532 (I2C), 16x2 LCD (4-bit), active buzzer. Wiring is in arduino_user_interface/arduino_user_interface.ino.
• Library: PN532 (e.g. PN532 + PN532_I2C in Arduino/libraries).
• Upload: Open the .ino in Arduino IDE, select board and port, upload.
• Serial: 115200 baud; commands READ or WRITE|userid; hub prints User ID when a tag is read.

In the web dashboard, use Connect to attach the hub via the browser’s Web Serial API (port is remembered in cookies). Tap a bracelet; the dashboard gets the User ID and fetches patient data and AI/risk results from the API.

Roughly 370,000 ED patients experience serious harm from diagnostic errors each year. Early decisions by EMTs and nurses can be life-or-death; lack of prior patient information leads to bias and misdiagnosis. TriageID aims to give emergency staff ready, rapid access to curated, at-risk patient information and AI-assisted insights.

Planned improvements from feedback:

• Run a local LLM for AI overview (instead of external API) for health security.
• Apple Watch / WearOS integration (NFC).
• Integration with NetCare (Alberta Health platform).
• Support multiple NFC bracelets per patient (e.g. if one is lost).

• DevPost DevPost Link
• Video Demo Video Demo
• Pitch deck: TriageID Pitch Deck

• Megh Mistry
• Kulgagan Bajwa
• Hasish Karri
• Hari Mallampalli