Inspiration
Arizona is preparing for a statewide participatory surveillance system where people can self-report symptoms, exposures, travel, animal contact, and environmental concerns. We wanted to build the intelligence layer around that idea: something that can spot weak signals early without diagnosing people, blaming places, or creating panic.
What it does
SpotSignal AZ turns incoming self-reported health data into explainable risk profiles for individuals and public health reviewers. It combines symptoms, optional image context, recent place/exposure context, weather/vector conditions, Epydemix Arizona contact-pattern data, and synthetic community trends. The system generates a personal signal, a GeoSignal map, reviewer dashboard, Signal Trace, Gemma triage/audit, and a human review queue.
How we built it
We built a React/Vite prototype with TypeScript. The app uses a local risk engine, mock Pima County trend data, Leaflet/OpenStreetMap for the map, NOAA/NWS weather context, Epydemix public GitHub data, Gemini for image/risk explanations, and Gemma for surveillance pattern triage and signal audit. The demo intake form simulates incoming reports from Arizona’s planned self-reporting system.
Challenges we ran into
The biggest challenge was balancing usefulness with responsibility. We had to make the system feel powerful without implying diagnosis or causation. We also had to clarify that SpotSignal is not replacing Arizona’s planned self-reporting system; it is the AI context layer around it. API key access and model availability also required mock fallbacks so the demo stays reliable.
Accomplishments that we're proud of
We built an end-to-end prototype: simulated intake, personal result, AI pattern triage, Signal Trace, real map, dashboard, review queue, CalmConnect, model card, and Arizona-specific contacts/resources. We are proud that the system is privacy-aware, non-diagnostic, human-in-the-loop, and directly aligned with the challenge requirements.
What we learned
We learned that early outbreak detection is not just about scoring symptoms. It requires context, uncertainty, trust, privacy, and human judgment. We also learned that a good surveillance tool has to be calm and explainable because scared users will stop reporting or misinterpret results.
What's next for SpotSignal
Next, SpotSignal would connect to a real Arizona self-reporting feed, verified reviewer accounts, EpiCore-style validation, clinic aggregate trends, vector surveillance, animal/environment partner data, and stronger multilingual support. We would also move AI calls behind a secure backend, improve model evaluation, and test the system with Arizona public health and community partners.
Log in or sign up for Devpost to join the conversation.