Patho: COVID-19 Forecasting Across 3,142 U.S. Counties
LLM-driven epidemic forecasts through 2028 across every U.S. county
A full-stack epidemic intelligence dashboard built at WashU’s Google DevFest, where it won Best Use of Featherless AI out of 72 teams. The frontend is a Next.js app using react-simple-maps to render a clickable, zoomable map of the United States that drills from a national view down to individual counties, with case and death metrics rendered through dynamic color scaling. The FastAPI backend filters nationwide CDC datasets down to specific FIPS county codes on demand using Pandas, with stateless request handling so the API stays fast even on large queries.
The interesting technical work is on the accessibility side. Traditional data dashboards are deeply visual, and standard alt text fails to capture dynamic geographic trends. Patho’s roadmap pipes regional data through Meta-Llama-3 via the Featherless API to draft conversational summaries of whatever county the user has selected, then streams the audio response through ElevenLabs so screen reader users get something genuinely informative instead of coordinate readouts. The current build serves data for all 3,142 U.S. counties with the predictive forecasting pipeline architected and ready to operationalize.