We worked on a project for the elderly: we wanted to detect anomalies in their behavior to scale social workers and draw their attention to the seniors who actually needed assistance. Is there an anomaly in their behavior (lack of heart-rate variability is thought to be a sign of depression or other problems? are they walking less than usual? are they sleeping well? what is the proxy for loneliness? what is a proxy for anxiousness? are they waking up in the middle of the night when they didn't use to? Have they fell?)

The mobile application was for social workers to track the behavior of seniors who were assigned to them. They received alerts based on some triggers and rules after analyzing the data that we consulted cariologists, social workers, paramedics for.

It was psychologically hard because the question I always asked the team was "Would you trust your parents' life with this?", and we built against that standard with the constraints at the time with the answer almost always being "no, I can improve X, Y, and Z". It was never enough.