Opportunistic service composition is a novel and disruptive approach for building software in dynamic and open pervasive environments. It aims to tackle the growing complexity of software design in such environments by dynamically providing relevant applications in the absence of explicit user needs: an intelligent engine composes software components that are present in the pervasive environment in order to build usertailored context-adapted applications, relying on reinforcement learning. This demonstration presents the current status of our opportunistic composition prototype: the intelligent engine builds pervasive applications in bottom-up mode from actual software components that are discovered via the UPnP (Universal Plug and Play) protocol, taking into account learned user preferences.