How rate-of-change detection makes a range hood 10x faster to respond

Gas stoves emit real pollutants. The mitigation is simple: turn on your hood when you cook. The problem is that most people forget. I have two small kids. I read too many air quality studies. So I automated my range hood.

Most smart hood automations watch for a temperature threshold: “when the kitchen reaches 26 degrees, turn on the hood.” Two problems. It’s slow. By the time the kitchen heats up, you’ve been cooking for 5 to 10 minutes, breathing the pollutants the hood was supposed to remove. And on a hot summer day the kitchen is already at 28 degrees, so the system either fires without a stove or needs seasonal recalibration.

When you turn on a burner, the air near the stove starts rising in temperature immediately. The derivative of the reading over a rolling two-minute window spikes almost as soon as the pan heats. When you stop cooking, the rate flattens and the hood shuts off. Hot days don’t cause false triggers because the temperature is high but not changing fast.

The product

The automation watches both temperature and humidity rate of change. Temperature catches most cooking. Humidity catches the boiling-water case, where steam arrives before heat.

UI-adjustable thresholds. Every sensitivity setting is exposed as a dashboard slider. A user can tune behavior without editing YAML. This took the project from “works for me” to “works in someone else’s kitchen.”

Manual override detection. Toggle the hood by hand and the automation steps back for 30 minutes. The system helps; it doesn’t fight the person holding the pan.

Safety shutoff. After two hours of continuous operation, the hood forces off and posts a notification.

Total cost is about $85: a Zigbee sensor ($15) and a SwitchBot Bot plus Hub Mini ($70). No cloud, no subscriptions, no rewiring. Everything runs on the local Home Assistant instance.

I described the rate-of-change idea to Claude Code and it generated the YAML scaffold. I read every line, pushed back on the over-engineered parts, and deployed. The thresholds needed a human and a stove. I ran real cooking sessions and tuned until false triggers hit near zero.

The automation has held up through more than a year of daily cooking. The repo is public under MIT.