This paper presents an approach to data mining based on an architecture that uses two kinds of neural network-based agents: (i) an instantaneously-trained surface learning agent that quickly adapts to new modes of operation; and, (ii) a deep learning agent that is very accurate within a specific regime of operation. The two agents perform complementary functions that improve the overall performance. The performance of the hybrid architecture has been compared with that of a back propagation network for a variety of classification problems and found to be superior based on the RMS error criterion.