|Abstract||The DJ career often involves to acquire music knowledge and to learn to use turntable techniques in order to express this knowledge on a DJ performance mix. In this thesis, we propose a prototype of a DJ-support Agent that takes advantage of Audio Mosaicing (AM), beat-tracking and data mining algorithms in order to help the DJ in the development of his/her mix. The prototype implements the basic functions of a DJing metaphor system such as, loop, cue, tempo, volume, crossfade, etc, under a tangible user interface. In addition it may enhance the nowadays DJ's song interaction in three different ways. First, the system suggests a surrogate similar song to the master song which is being played back. This surrogate song is built from song slices taken from the rest of songs inside a music collection. Second, the user has control over the parameters of similarity, therefore the user may experience, in real time, the process of building up the suggested surrogate song. Third, at any time while a master song is being played back, the system is able to synchronise multiple songs to this master. The last allows the user to merge the mix, coherently and flexibly, into a new selected song. In order to implement these functionalities first, the song collection is cut into quarter-note slices (candidates), the framework retrieves several Low Level Descriptors (LLD) for these candidates in an o-line analysis. The real-time prototype's core consists of a continuous comparison between the LLD already stored and the ones from a master/reference song which is being played. We explore similarity in a bidirectional fashion. On one hand, the system takes the master song as the reference and proposes existing similar candidates. This allows the proposed candidates to build up a new version of the master song in which the rhythm is maintained and the control of similarity is left to the user. On the other hand, we let the system change the source of reference to the last proposed candidate. The prototype lets the user access the original song where the candidate was taken, and start playing it from this position. The overall performance of the user is increased since it is able to build up coherent mixes in a rapid and flexible way. The prototype ts to skilled and non-skilled DJs since the implicit knowledge one uses to mix is adopted by the core of the system. This system has been evaluated with a subjective test over 15 subjects. The test has reflected no statistical significance over the user ratings on the system performance.