This paper provides a brief introduction to our experiences and results of applying artificial intelligence techniques to the real world problem of predicting inflow to a Sewage Treatment Plant. The basic structure of the plant is described followed by a discussion of how recurrent neural networks may be applied to the problem of predicting systems with diurnal and chaotic components. Next, the software system developed and implemented on the site is presented. A brief discussion on the feasibility of incorporating the prediction module within a hybrid system environment follows. Finally, the possibility of using fuzzy logic for control in similar problems is discussed with reference to some preliminary experiments.