Skip to Main Content

Lusong

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Lusong — Demo video

Lusong answers a Filipino motorbike commuter's morning question: should I leave now, and which way? Type a trip anywhere in the Philippines into From and To boxes, as in a maps app, or write it the way you'd say it, like "Talamban to IT Park, leaving around 7". Lusong gives one of four answers: Go, Go another way, Wait, or No one knows. The map is the page. It shows PAGASA's advisory areas, flooded spots other riders reported, road closures found on the web, and rain now. Every answer lists its evidence, with the age of each item and where it is. Anyone at the water can send a photo or a voice note in Bisaya, Tagalog or English, from the page or through the Telegram bot. Bisaya notes were tested live, and English and Tagalog notes on synthetic audio: English 4 of 4, Tagalog 2 of 4, 2 draws each. Lusong measures the water against the MMDA flood gauge that Filipinos already know. A spot read as "not passable to light vehicles" is blocked for three hours. Lusong then asks the router for a way around it and checks that way against the reports too. Under an advisory or rain it says Go only when clear photos cover the whole route. Otherwise it says No one knows, because nobody has looked yet. We tested the depth reader on 45 openly licensed photos whose captions name a depth, 2 draws each. On the 32 flood scenes, its deciding reading matched the caption on 28 of 64 draws and read too shallow on 22. That's too shallow too often, so the rules lean safe. A deep reading blocks a spot, and only a photo taken within the hour that shows no flood at all can clear one. A voice note never clears a spot. Lusong can also watch a trip. Pick a morning window, and it checks the trip every 10 minutes and messages you on Telegram when the window opens and whenever the answer changes. A PAGASA advisory, road closure or disaster-office post that touches your trip is sent to you once, whatever the window. Nationwide ones go to every rider.

Team