Social media sites are continually evolving as platforms of social exchange, interpersonal communication, and psychological expression. In this talk I will present how this rich, longitudinal repository of data can provide a new, disruptive mechanism to assess and understand our mental well-being. By computationally measuring linguistic and behavioral shifts on social media across major life events like childbirth, and personal and collective trauma, I will discuss how we can identify risk markers of a variety of mental health challenges: postpartum depression, posttraumatic anxiety, and suicidal ideation. I will also show how we can develop machine learning based, causal, multimodal, and mixed-initiative predictive models that can infer an individual’s or an online community’s forthcoming mental health risk, and its trajectory over time. Finally, I will discuss the technological, clinical, social and ethical implications of this body of work in improving mental health, ranging from enabling early detection and awareness, to designing interventions that can bring timely, personalized help and support to vulnerable populations.