For most people in Africa, the flood warning is the water itself.

Forewarn warns you days before a flood reaches your area: your street, your farm, your village. Free, anywhere in Africa, no account needed.

We replayed the forecasts of the week before the Nairobi flood of 23 April 2024 through Forewarn. It would have flagged flooding as possible 7 days before the water came. See the replay.

How Forewarn warns you

• Ask in plain words. Put a pin on your door and ask if it will flood. The answer gives a level (unlikely, possible or likely) and the chance in words, such as "about 1 in 4 forecasts".
• Speak your question. Tap the microphone and ask out loud in English, French or Swahili. Meta's Muse Voice Transcribe writes it down, and you can fix the words before you send. Each language was tested on recordings of native speakers before it was turned on. Answers can be read aloud, in Swahili, Yoruba and Hausa with Meta's MMS voices.
• 14 days ahead. 51 runs of the European (ECMWF) forecast for the spot you pick, counted for you.
• How wet the ground is. The last 30 days of rain against 30 years of records. Soaked ground floods sooner.
• The last flood, on the map. Lagos and Nairobi, street by street: the last storm's water over the satellite map, hour by hour.
• Send a photo of the water. Meta's SAM 3.1 marks the water and the people, Muse Spark says what the water reaches, and people are pixelated.
• Alerts a person approves. An agent checks your area every 6 hours and drafts an email alert when flooding looks possible. A person approves it before it goes out.
• Warn your family and neighbours. Share a warning on WhatsApp as text, an image card, or a voice note in Swahili, Yoruba or Hausa.

Built on Meta's models

• Muse Spark Open: the agent. It reads the forecast and the ground, moves the map, and writes answers, alerts and family messages.
• Muse Spark 1.3: reads a flood photo: what the water reaches, whether it moves, what to do.
• SAM 3.1: finds the water and the people in a photo, so people are pixelated.
• Muse Voice Transcribe: turns a spoken question into text, in English, French and Swahili.
• MMS voices: read answers aloud in Swahili, Yoruba and Hausa.
• Omnilingual ASR: Meta's open speech model; in our test it understood Yoruba.

Every number has a source

Code computes every number, and a checker compares each number the agent says with the data. The agent runs on Meta's Muse Spark Open.

We replayed 36 past African floods with today's rules: 29 were flagged in advance (81%), 28 of them 3 days or more ahead. See the replays in the time machine.

Forewarn is not an official warning service. If your local officials say leave, leave.

The code: github.com/euphoriawolf/forewarn.

For AI tools: llms.txt and the full description.

Built by Derrick Ikenga for the Meta Global AI Developer Hackathon.