There is growing need for automated tools to reason about groups in diverse countries. For example, the CDC may wish to understand social patterns that cause diarrheal diseases to spread in Kenya, the World Bank may want to understand how various socio-economic-religious factors may affect the actions of tribes in the Pakistan-Afghanistan border region, while domestic and international security agencies may want to track Kurdish organizations that share characteristics with other organizations that have resorted to violence in the past . In this paper, we describe the CARA architecture for gathering data about different cultural groups, learning the intensity of opinions that those groups have on various topics, and developing a process that supports building/extracting models of behavior of those groups and continuously refining those models through shared, multi-person, learning experiences. The CARA architecture is supported via ongoing applications we have been developing - one in global health care, another in tracking tribes along the Pakistan-Afghanistan border with a view to understanding and eventually reducing the burgeoning drug trade there, and a third focusing on politically active minorities at risk in fragile regions of the world.