Inspiration
In Malawi, a mother fills a jerrycan from a borehole she trusts. Three days later her child has cholera. The lab result that could have warned her arrived after the funeral. We built FWS so the warning comes BEFORE the sickness, from data that already exists: satellites, rainfall, field records.
What it does
FWS watches drinking water and flags risk early. It fuses satellite water-color signals, cholera history, rainfall forecasts, and lab data into one risk picture, then tells responders the single water point that matters most, and the verified-safe alternative nearby, in SMS and prerecorded Chichewa. One warning can reroute a village before a single child drinks bad water.
Track Alignment, Track 6: Resilience Informatics
FWS is an early-warning and resilience system: it predicts contamination risk before it spreads, alerts responders, and names the ONE water point to act on first, exactly what resilience informatics is for.
How we built it
A modular One Health pipeline: Sentinel-2 water-color and turbidity signals, WPdx water-point registry and lab data, rainfall and cholera forecasting, an unseen-anomaly detector, and a dependency network that traces how contamination propagates. Backend in Python/FastAPI with PostgreSQL; dashboard in Next.js with a UI grounded in real satellite and spectral imagery.
Challenges we ran into
Evidence is thinnest where it matters most, when data is scarce, FWS says FIELD SAMPLE REQUIRED instead of guessing. Water points don't exist alone, so we model how contamination spreads through communities. And we were honest about scope: EYES, BRAIN, DECISION, and VOICE are demoed and working; NETWORK, HANDS, VERIFY, and MEMORY 2.0 are designed and staged for field data. We'd rather show you exactly what works than pretend everything does.
Accomplishments we're proud of
Ten integrated modules, from satellite eyes to field verification to outcome feedback, that turn scattered data into one prioritized action, in the language a community actually speaks.
What we learned
An early warning only saves a life if it reaches a phone in Chichewa. The best model is useless in a village that can't read its alert. And honest uncertainty beats false confidence every time.
What's next for Fresh Water Sentinel
Closing the loop: verification outcomes feed back into risk scoring (MEMORY 2.0), wider water-point coverage, and field trials with local health offices, so the mother fills her jerrycan somewhere the data already says it's safe
Log in or sign up for Devpost to join the conversation.