ERLyAlert

Inspiration

The idea was born out of a simple but urgent problem: delays in emergency medical response can mean the difference between life and death. We wanted to create a system that allows individuals to get help faster.

What it does

It forecasts emergency room visit surges up to 14 days in advance, helping hospitals prepare staff and resources. It alerts users when high patient volumes are expected and explains the key drivers behind these surges—like severe weather, public events, or air quality issues. Healthcare staff can interact with a visual dashboard and even ask questions via chatbot.

How we built it

We built the backend using FastAPI and integrated a hybrid forecasting model combining Facebook Prophet and LSTM. Prophet handles trends and seasonality, while LSTM detects complex patterns in time-series data. We used Pandas, NumPy, and scikit-learn for data processing. The frontend was developed with Streamlit, with visualizations powered by Plotly and Pydeck. Real-time external data is securely pulled using Python scripts and APIs.

Challenges we ran into

Aligning ER visit logs with external datasets that used different timestamps, timezones, or granularity. Finding the right balance between responsiveness and stability in forecasts. Streamlit needed optimization to handle interactive features like map overlays and multi-region comparisons without lag. Making risk drivers transparent and interpretable, especially when multiple external factors were involved.

Accomplishments that we're proud of

Deployed a functioning multi-region forecast engine with real-time retraining. Created a visually compelling dashboard that gives both overview and detailed insights. Integrated external data sources that meaningfully enhanced prediction quality. Delivered contextual alerts that don’t just say when, but why risk is elevated.

What we learned

Real-world data is messy—but worth the effort to accommodate to. Temporal modeling is powerful, but contextual awareness from external features makes it exponentially more useful. Usability matters too as dashboards must be clear even when the underlying system is complex. Automation is important as well as retraining, alerts, and data pipelines must be as hands-off as possible for real-world adoption.

What's next for ERLyAlert

Some of our goals are to: Work with a local health department or hospital to trial the system in the field. Add intra-day predictions or hour-level spikes. Notification system: Implement real-time email/SMS alerts with thresholds. Allow annotations, comments, or overrides by health officials. Improve responsive layout for tablets and phones.