With a growing list of sensing modalities and AI models, the potential of wearable devices in mental health care continues to grow. I will present several research projects that combine novel data and AI in this space. First, Electrodermal activity (EDA) is a physiological measure that is used to index sympathetic arousal in response to stressors and other perturbations. The Fitbit Sense 2 and Pixel Watch 2/3 are the first commodity hardware devices with continuous electrodermal sensors, I will present an analysis of over 10 million hours of continuous EDA data collected from over 16,000 people.  I will then show differences in daily EDA between those with high versus low depressive symptoms. Next, I will show the promise of scaling foundation models for these data that could create generalizable models that are label efficient. Finally, I’ll show results of how these types of sensor data can be combined with large language models (LLMs) for zero-shot inferences about psychological functioning.