National AI Challenge Winners · A finished product in just 2 Weeks
Safe WatersA predictive bathing-water advisory system, conceived, built, and deployed to production by a Presage-led team — most voted winners selected from about 1,000 participants and 200+ teams.
Lead time gained vs. the next scheduled lab sample
Reliable — uncertainty matched reality during backtesting
CO₂ saved per season across 10 sites
From first commit to a live, production-deployed app
The Problem
Local authorities are required to monitor and warn the public about bathing water quality at over 240 official sites nationally. But water samples sent to a lab take 1–2 days to come back, and beaches are typically tested only monthly or fortnightly.
That gap means contamination, often triggered by environmental factors such as heavy rain, can go undetected for days: people swim in unsafe water before an advisory is even possible, or sometime a "prior warning" advisory which was issued often stays in place longer than necessary, costing tourism and water-sports revenue.
Testing more frequently is a cost, logistics, and coordination problem. More samples mean more collection trips, more lab fees, and more officer time spent checking and publishing advisories.
The Solution
Presage and Safe Waters aim to solve similar problems: predicting the future, whether that’s an unexpected fault in industrial machinery or bathing water quality seven days ahead.
Using environmental data — and with no new sampling infrastructure required to start, we deliver a forecasting solution which captures the water quality 97% of the time.
Explainable
Each site shows which environmental factors are pushing risk up or down that day — so an officer can see why the model predicts what it predicts.
Reliable
A 7-day forecast with both data and model uncertainty shown separately, against the regulatory threshold. The interval was checked against 10+ years of held-out historical data before it was trusted with real forecasts.
Officer-Reviewed
When risk crosses a threshold, Safe Waters drafts the decision for the officer: approve, override in either direction, or request a lab sample. Nothing is published without sign-off.
Impact
Figures reflect potential savings for a 10-site deployment, based on stated assumptions on review time and collection logistics.
Built, Not Just Prototyped
React frontend, .NET API, SQL Server database, running end-to-end and demoed live over a public link.
Production-ready architecture: containerized services (Docker Compose), JWT authentication, role-based access control.
Built for multiple councils from day one — a Reviewer can be scoped to a single county, so one deployment serves many local authorities without cross-visibility.
Automated test coverage across the forecasting model, the API, and the UI.
The Team
Responsible by Design
Every forecast comes with the key drivers behind it, in plain language — e.g. heavy rainfall tomorrow will likely raise E. coli contamination.
Shows a confidence band instead of false precision. In back-testing, that uncertainty matched reality 97% of the time across 10 real beaches and 10+ years of data.
Can auto-draft an advisory notice for an officer to review and approve in one click. The system never issues a warning on its own.
The Recognition
Winners, 2026Selected as most voted winners from about 1,000 participants and 200+ teams, for a system built and deployed in two weeks.
