Multimodal conversational dialogue systems consisting of numerous software components create several challenges for the underlying software architecture and development practices. Typically, such systems are built on separate, often pre-existing components developed by different organizations and integrated together in a highly iterative way. Here, the traditional dialogue system pipeline is not flexible enough to address the needs of highly interactive systems, such as parallel processing of multimodal input and output. We present an architectural solu-tion for a multimodal conversational social dialogue system developed in a large international project.