When you put sensors in multiple homes and watch them for a month, you stop thinking about motion detection. You start thinking about people.

This month I've been watching three households — my apartment in California, and two homes in Bangalore. And the thing that keeps striking me is how completely different a "normal day" looks depending on who's living it

The couple here is in their 30s, both working. Which means from 9am to 6pm, the home is a flatline. Nothing. Total silence.

Then 6pm hits and the house comes alive. Cooking, TV, movement between rooms. A completely different signature than anything I'm seeing in Bangalore, where someone is almost always home.

What this taught me: "no movement for four hours" means something completely different depending on whose home you're in. For this couple, it's Tuesday. For an 80-year-old living alone, it might be an emergency.

The system has to know the difference.

Here the range is 55 to 80+. And these homes never really go quiet — someone is almost always pottering around. Morning coffee, afternoon rest, evening family-time. The rhythm is slower, more continuous, more present.

But the thing that surprised me most wasn't daytime. It was night.

Older people move a lot in their sleep. The sensors are sensitive enough to catch the smallest shift, and I was not prepared for how much data that generates overnight. What looks like "activity" to the system is just someone turning over. Several times an hour, sometimes.

For a while I thought something was wrong with the sensors. Then I realized: this is just what it looks like when an 80-year-old sleeps. And if I'm not careful, the system will spend all night convinced something is happening when nothing is.

The core insight this month is that I'm not building one product. I'm building a system that has to understand this person, in this home, with this particular version of normal.

A 30-year-old DINK household and an 80-year-old in Bangalore have almost nothing in common behaviorally. Same sensors. Completely different signal.

Getting that right — learning each person's specific baseline instead of applying generic thresholds — is the whole game. And I'm starting to see what that actually means in practice.

Next up: correctly identifying wake time across these very different homes. Also waiting on iOS beta approval so real users can finally get the app in their hands.