A self-organising, peer-to-peer, network model, iXChange, is presented. Peers share information, such as usage history and centres of interest, to self organise into semantically clustered groups and intelligently elect super peers using a neural network. Super peers can efficiently route queries based on a small world model and demonstrate a strong self organisational capability using ontology based interest representations. Preliminary results show that the neural network super peer election architecture can be trained in an acceptably low amount of time resulting in successful autonomous election of efficient and reliable super peers.