EmberGuard

Inspiration

Wildfire risk models are often locked behind academic paywalls or complex technical dashboards. We were inspired by the local stakeholders—farmers, park rangers, and homeowners—who are the first to respond but the last to receive actionable data. EmberGuard was born from the desire to democratize wildfire intelligence. We took the power of ML-driven risk modeling and stripped away the noise, delivering a "snappy," user-centric experience that turns "big data" into "local action."

What it does

EmberGuard is an end-to-end wildfire intelligence platform that translates raw environmental data into personalized safety narratives for landowners. By combining broad national weather forecasts with "ground-truth" data from local sensor hardware, the system monitors the specific conditions of a user’s property in real-time. This information is processed into easy-to-understand risk reports that explain exactly how current heat, humidity, and air quality levels affect local ignition threats. Users can explore their land through an interactive map and a timescale slider, allowing them to visualize how changing weather patterns shift their safety profile over time. Ultimately, the platform strips away technical complexity to provide a clear, actionable story of a property's vulnerability.

How we built it

EmberGuard's web app is built on a Next.js and Tailwind frontend with Mapbox for spatial visualization. A FastAPI backend orchestrates the flow of data between a Supabase PostgreSQL database for data storage, an NWS API endpoint for live weather conditions, and a Puter.js interfaced Claude API endpoint powering the risk modeling and report generation. Our XGBoost classifier is hosted on Render. EmberGuard's field units are equipped with an Arduino Uno R4 Wifi, a temperature sensor, a moisture sensor, and an air quality sensor for continuous real-time data that is fed to the web app.

Challenges we ran into

Starting out, we had difficulty navigating the interface between hardware and software, as getting the Arduino to send data over wifi proved to be more challenging than expected. Training our XGBoost classifier also presented various design challenges, such as choosing which dataset to train on, finding which model performed the best on our chosen dataset, and finally tweaking model hyperparameters to optimize the model's performance metrics.

Accomplishments that we're proud of

Given the team's overall experience level, we are very proud to have finished a full-stack project in 24 hours. This was a first-time hackathon experience for two out of our three members.

What we learned

Through building EmberGuard, we learned the importance of planning things out step by step, and taking tasks one at a time. Our project is the culmination of one small iteration after another, each one building towards our finished product.