Flood early warning for the Philippines

Give communities
more time.

HydroSentinel watches Philippine rivers, typhoons and rainfall around the clock. A Muse Spark agent turns the live data into a verified briefing and alerts people can read in their own language.

Built with Meta's Muse Spark, SAM 3.1 and Muse Image through the Meta Model API

Right now

Loading live figures…

Is my river safe right now?

One answer, in plain words.

The river closest to its own watch level, its last week and the forecast, the nearest storm, and the SMS residents would receive in their language. Every number comes from the live data and is checked by our server before it is shown.

See all 17 basins

Loading live river data…

Why it matters

When the water
moves, context matters.

An average of 20 tropical cyclones enter the Philippine Area of Responsibility each year, according to PAGASA. Flood response is a handoff problem: from a sensor to a forecast, from a forecast to a family, from uncertainty to a decision.

01

See the signal

Live river gauges, a 17-basin Philippine outlook, rainfall and typhoon feeds, each with its source.

02

Predict the shift

A backtested forecasting engine turns noisy readings into a view of what comes next, with honest uncertainty.

03

Move sooner

A Muse Spark agent writes a cited, verified briefing and SMS alerts in five Philippine languages.

Who it helps

Built for the people closest to the river.

Barangay and municipal DRRM offices

A plain-language briefing on which rivers are rising, how close each is to its watch level and what to do next, with an SMS ready to forward in the language residents speak.

Provincial and LGU planners

A 30-day flow outlook for 17 major basins, from the Cagayan in the north to the Mindanao River in the south, with alert levels based on each river's own five-year record.

Families in flood-prone areas

Short, calm messages in Filipino, Cebuano, Ilocano, Hiligaynon or English, that always point to PAGASA and local authorities for official warnings.

Meta AI inside

Four jobs, three Meta models.

Each model does work a person would otherwise have to do by hand, and each result is labelled and checkable.

Muse Spark

Flood-watch agent

Decides which of six tools to call (hazard feed, river basins, rain forecast, gauges), cites a source for every claim, and drafts SMS alerts in five Philippine languages. Our server checks every number against its source before anyone sees it.

Runs every three hours, or on demand.

Watch it investigate

Muse Spark

Ask in plain language

Answers questions about the rivers with structured output: what was measured, what is forecast, how sure we are and what to do. It says so when the data can't answer, instead of guessing.

Grounded only in the live readings and outlook it is given.

Ask a question

SAM 3.1

Flood photo check

Outlines the floodwater in a photo from the field and reports how much of the scene is under water, so reports from the ground can be compared over time.

Photos go to SAM 3.1 for analysis and are not stored by HydroSentinel.

Try a photo

Muse Image

Community posters

Draws a preparedness illustration matched to today's alert level, to send alongside the SMS so the message also works for people who read less.

Labelled as AI-generated wherever it appears.

Make a poster
What you get

Everything runs on live data.

Verifiable agent

Muse Spark calls tools, cites evidence, and the server checks every number.

17 Philippine basins

GloFAS river flow with a 30-day forecast and alert levels based on each river's own history.

Typhoon feed

GDACS cyclones and floods, ranked by distance to the Philippines.

Measured gauges

Real river levels every 15 minutes test the forecasting engine against reality.

Five languages

SMS alerts in English, Filipino, Cebuano, Ilocano and Hiligaynon.

5 languages

Backtested forecasts

Each forecast is scored against a simple baseline on data the model never saw.

Photo check

SAM 3.1 measures how much of a field photo is flooded.

Community posters

Muse Image illustrations matched to today's alert level.

Rain outlook

Open-Meteo rainfall, hour by hour, for every gauge and basin.

Honest by design

Models are labelled as models. Official warnings stay with PAGASA.

Works on any phone

Installs like an app and keeps the last data when the signal drops.

Open and exportable

Every reading has a source, and every series exports to CSV.

Built to be trusted

A warning is only useful if people believe it.

Every number is checked

The agent must cite the tool result behind each claim. Our server compares every number against that source and flags any that don't match, before the briefing is shown.

Forecasts are tested first

Three models compete for each forecast horizon and are scored on data they never saw, against a simple no-change baseline. The dashboard shows the scores.

Open data with a trail

GloFAS river flow via Open-Meteo, GDACS disaster alerts, UK Environment Agency gauges and Open-Meteo rainfall. Every reading keeps its source and exports to CSV.

Secure by default

Model keys never reach the browser. Changing alert levels needs an admin key, AI features are rate-limited, and every page sends strict security headers.

How it works

Intelligence that
earns trust.

Every alert keeps a trail: the data, the model, the uncertainty, the evidence it cites, and the reason it matters now.

Watch the agent work
Try it yourself

Four things to try in five minutes.

  1. Watch the agent investigate

    Open the Agent tab and follow each tool call, then tap any citation to see the exact data behind it.

  2. Check any Philippine river

    Pick a basin on the map to see its 30-day flow, its alert levels and today's status.

  3. Test the data checker

    Download our demo file, a synthetic typhoon at a Marikina gauge with real-world gaps, and run it through the validator.

  4. Ask in your language

    Try a question in English or Filipino, such as “Ligtas ba ang Ilog Cagayan ngayong linggo?”

H
Ready when the next storm is

Let's make the
next hour count.

Enter HydroSentinel