Breezy: AI-Powered Asthma Risk Detector

Introduction page that provides the option of seeing the demo or creating your own page
UI/UX resembling Pokemon go to reveal nearby envrionmental stimulants of asthma

Inspiration

Our inspiration for Breezy came directly from our teammate Alexis, who lives with asthma. Managing asthma daily isn't just about carrying an inhaler, it's a constant effort to anticipate unpredictable environmental triggers like pollen spikes, sudden weather changes, and air quality drops. Watching Alexis navigate these daily challenges made us realize how few tools exist that proactively predict asthma risks before an attack occurs. We wanted to build a smart, accessible solution that empowers individuals with asthma to plan their days safely and breathe easier.

What it does

Breezy is an AI-powered asthma risk detector that analyzes environmental factors, real-time weather metrics, and user-reported personal health indicators to predict potential asthma flare-ups. By synthesizing environmental data with predictive modeling, Breezy alerts users to high-risk conditions before they become critical, offering actionable insights and personalized safety tips.

How we built it

Frontend: Built with a responsive interface designed to provide quick, intuitive risk assessments at a glance.
Backend & AI/ML: Developed using custom logic and machine learning models trained on environmental data (air quality index, humidity, temperature, and allergen counts) to evaluate individual risk scores.
Integrations: Embedded real-time data feeds to continually track local climate conditions.

Challenges we faced

Data Integration: Aggregating and normalizing real-time environmental data alongside personal health factors required precise calibration to prevent false alarms.
Balancing Accuracy and UX: Fine-tuning the predictive model so that risk factors are communicated clearly and calmly without overwhelming the user.

Accomplishments that we're proud of

Transforming a personal experience into a fully functional, real-world application that directly addresses a major health challenge.
Building a reliable predictive pipeline that converts complex environmental data into clear, actionable risk levels.

What we learned

The critical importance of user-centered design in healthcare and accessibility tools.
How to effectively process environmental data inputs for real-time predictive analytics.

What's next for Breezy

Integrating real-time wearable device telemetry (heart rate, respiration rate) for higher precision.
Expanding push notifications for dynamic geo-fenced safety alerts when entering areas with elevated environmental risk factors.

Built With

fastapi
gemini
github
google-colab
joblib
json
lightgbm
mongodb
numpy
open-meteo
pandas
python
react
scikit-learn
shap

Submitted to

ShellHacks 2026

Created by

I built Breezy's full ML pipeline end-to-end: merged real Open-Meteo weather data with synthetic patient logs to create a labeled training dataset, engineered a no-leakage flare-up label plus 14 predictive features (rolling averages, personal-baseline deviations), and trained a LightGBM classifier (0.79 recall, 0.87 AUPRC) tuned to prioritize catching real flare-ups. I added a SHAP-based explainability layer generating per-user "trigger profiles," then shipped the whole thing through a FastAPI backend with a saved model and JSON output schema for real-time frontend integration.

Emily Forestier
I built the React frontend for Breezy: a map home screen with a companion character, the forecast screen, the daily check-in quest with streaks, trigger patterns, onboarding and the emergency screen. I also made it installable as a phone app. On the backend, I added caching with a fallback so the demo survives API outages, set up MongoDB Atlas, and fixed our Gemini integration, which had been silently falling back to template text because of a retired model, a timeout that was too short and cut-off replies. Finally, I connected the frontend to the live backend.

alexislanier0420-tech
Junia Celestin