Spark AZ

About The Project

Arizona Health Guardian, also called Spark AZ, was inspired by a simple public-health problem: early outbreak signals are often scattered across many places. A person may notice symptoms, another person may report sick animals, weather may increase mosquito risk, and travel patterns may bring in new disease threats. Individually, these signals may look small. Together, they can reveal an early warning.

I built Spark AZ as a One Health surveillance prototype for Arizona. The app combines human check-ins, animal observations, environmental signals, weather, air quality, travel context, and county-level trends into one explainable risk score.

The core score is shown from 0 to 100:

[ Risk = f(Human, Animal, Vector, Environment, Community) ]

The project also includes AI-generated insights, but the AI does not publish alerts by itself. Alerts go through a human-in-the-loop admin console, where an analyst can approve, edit, or reject them before they become public.

What I Learned

I learned how important explainability is in public-health AI. A score alone is not enough. Users and analysts need to know why the score changed, what signals contributed to it, and how confident the system is.

I also learned that a strong health-tech prototype should not only have a user-facing app. It also needs analyst workflows, data-source transparency, privacy awareness, and a clear path from demo data to real deployment.

How I Built It

I built the project with a React/Vite frontend, Supabase backend, edge functions, seeded Arizona demo data, county-level aggregation, and Gemini-powered insight generation. The app includes a resident check-in flow, dashboard, map, insights page, simulator, data sources page, Arizona playbook, and protected admin console.

The risk model combines deterministic scoring with explainability. Each factor becomes a visible driver, so the user can see what pushed the score up or down.

Challenges

The biggest challenge was balancing realism with hackathon feasibility. A real public-health system needs approved data partnerships, privacy review, clinical validation, and analyst governance. For this prototype, I focused on showing the full workflow safely using anonymous check-ins, open/public data sources, seeded demo data, and clear model limitations.

Another challenge was making the app useful for both residents and analysts without making the interface confusing. I solved this by keeping the shared app experience simple, while separating analyst-only actions into the admin console.