Previous research shows that diverse teams in background and skills can outperform homogeneous teams. However, people often prefer to work with others who are similar and familiar and fail to assemble teams with high diversity levels. We propose a team formation algorithm that suggests diverse based on individuals' social networks, allowing them to keep high familiarity levels. Our novel algorithm is based on the NSGA-II genetic optimization that splits students into well-connected and diverse teams within an organizational network. It optimizes measures of team communication cost and diversity in O(n^2) time. The optimization finds Pareto optimal solutions that optimize both metrics, returning teams that have both diversity in member attributes and previous connections between members. We tested the algorithm on real team formation data collected from the MyDreamTeam platform. The solutions provided by the algorithm are superior to the teams assembled by the students, in both diversity and communication cost measures.