Existing crowdsourcing database systems fail to support complex, collaborative or responsive crowd work. These systems implement human computation as independent tasks published online, and subsequently chosen by individual workers. Such pull model does not support worker collaboration and its expertise matching relies on workers’ subjective self-assessment. An extension to graph query languages combined with an enhanced database system components can express and facilitate social collaboration, sophisticated expert discovery and low-latency crowd work. In this paper we present such an extension, CRowdPQ, backed up by the database management system Crowdstore.