@mbarrbosa took the stage at our latest AI Show & Tell to talk about Antifogo, a system that detects wildfires from satellite data and predicts where they'll spread.
Very specific problem, but surprisingly portable lessons. Three that stuck with us 👇
🧠 Know how your inputs fail, not just what they say. Antifogo runs on satellites, weather, terrain and fuel. Each one breaks in its own way, so the job isn't picking the best source, it's knowing what each one's silence means. That's true for anything running on data you didn't collect yourself.
🧠 Keep what you measured apart from what you guessed. Blend the two and nobody can tell how much to trust the output. Antifogo labels every number as observed or inferred, and shows the fires it missed next to the ones it caught. You lose the tidier headline, and you get a claim that holds up when someone pushes on it.
🧠 Optimise the part of the chain that's actually yours. Marco's framing: the system doesn't control when firefighters arrive, only when the information reaches them. Knowing which link is yours is what stops a team promising an outcome it doesn't own.
Thanks, Marco, for walking us through the messy middle of the build. 🛠️