AURORA Lifeline

Know who a cyclone will cut off, before the roads close

Open Source

Artificial Intelligence

GitHub

Maps

Google Cloud Run

Visit WebsiteSee on Product HuntGithub ⧉Firebase ⧉

Hi Product Hunt 👋 Every cyclone season in India, warnings get better and evacuations save lives. But after landfall the roads flood, bridges and culverts close, and villages lose their only road to a hospital. A woman in labour, a snakebite, a dialysis patient: for days, help can't get through. Three days out, a district control room knows where the storm is going. What it doesn't know is which health centre will be cut off, how likely that is, when, and what to move there before the roads close. Today that's worked out by phone calls and experience. AURORA Lifeline starts from the official IMD bulletin and answers those four questions: • Gemini reads the bulletin PDF and quotes every value; code checks each one against the PDF. • Over a thousand storm futures from Google DeepMind WeatherNext and ECMWF, aligned to IMD's track, put wind, rain and surge on 5.4 lakh road segments, hour by hour. • For each future, a fast network search finds when every village and health facility loses its last road to a public hospital. You get a chance and a time window, never a falsely precise hour. • It ranks where to stage earthmovers, each with a deadline and its reason. • Gemini drafts the advisory in Telugu, Hindi or English, but it never writes a number: code inserts every figure. It's read aloud with Gemini TTS and exported as a CAP alert, and an officer approves before anything goes out. • An agent built with Google's Agent Development Kit answers officers' questions from the run's facts, with citations. Try the replay of Cyclone Montha (Andhra Pradesh, Oct 2025) exactly as it looked 62 hours before landfall, then Cyclone Dana in Odisha. Geography is configuration, so a new state is a config file and a map extract. How it evolved: we started with "which roads flood?" and ended with "which PHC loses its last road, and when?". That shift made us build our own road network instead of using routing APIs. It also made us strict about uncertainty: a chance and a window, labelled priors, and validation against Sentinel-1 radar with our misses published on the Proof page. It's open source (Apache-2.0) and was selected in the top 10 of Track 05 at Build with AI: Code for Communities. We'd love feedback, especially from anyone in disaster management or public health. What would make this useful in your district?