In the last decade, video distribution has evolved from analog broadcast to fully digital video. Today, users have the freedom to decide what to watch and when to watch it, using Internet video services. Recommendation and profiling systems are there to assist the user in creating a personalized content sequence. One step further, the content itself is personalized and enabled for interactive playout. Linear content such as daily news is analyzed, segmented, filtered, and reassembled per viewer based on his profile, preferences, and available time. The transport is typically performed in a unicast fashion from a central server, which places a high demand on the network since it prevents reuse. In this paper we introduce the paradigm of late assembly. Personalized and interactive videos are generated based on semantic segments and user profiles which are combined on intermediate nodes near the end user. Together with deadline scheduling and caching, an efficient transport is realized. An architecture and a prototype implementation of this system are presented. © 2011 Alcatel‐Lucent.