In this project we address the problem of automatic inference of relational social constructs in group conversations using nonverbal behaviour. We focus on dyadic social phenomena emerging in small groups - a novel area of research in social computing. Using probabilistic methods, NISHA aims to discover these dyadic nonverbal patterns, analyze their properties and use this knowledge to characterize relationships among the members of the teams. We work with audio-visual data from groups solving tasks together, recorded using portable sensors.