It has been attempted to develop a neurocomputing paradigm for artificial music composition in Indian style. A musical organization, comprising a melodic pattern, a rhythmic pattern and a melody violation part, has been considered for this purpose. A simple feed forward hybrid neural network with a functional link hidden layer has been so designed as to integrate such a musical organization with Indian raga rules. The proposed artificial neural network can output a variety of compositions on any raga by generating random link weights that shape the musical organization in various ways within the freedom available in the raga framework. The neural network can be customized with specifications of any raga for outputting a large variety of musical sequences that abide by the particular raga rules. In an interactive mode, a musician user may compose non-classical and light classical songs in Indian style based on desired ragas. An evaluation of such outputs from the system on Hindustani music, as done by a group of musicians, is encouraging. The present system, when run in an ''Auto Mode'', may also serve as a generator of ''semi-finished'' music for composing music of non-traditional structure.