Visualization pipelines and scene graphs are widely used paradigms in rendering applications, the former as a key concept in scientific visualization while the latter provides a powerful representation for 3D scenes in graphics applications. Despite an apparent disparity in usage, they show structural similarities by both using a graph-based architecture. Nevertheless, few attempts have been made to integrate both visualization pipelines and scene graphs in a unified design. This paper presents an architecture based on an extended dataflow network which allows implicit invocation of callbacks on network events. Events are then handled by programmable layers on top of the "dataflow layer" to dynamically build scene graph structures and define a multiparadigm model interface, which forms what we call an "indexed dataflow network". We show an implementation of our architecture using NVIDIA SceniX scene management engine and Qt's model/view framework.