Nowadays, commercial applications for expert system's generation (shells) provide the ability to implement knowledge-based systems simply by filling it with relevant knowledge for the system's area of application. The manipulation of that knowledge is usually done by pre-defined procedures made available in that shell. Nevertheless, the generic purpose of a shell normally disregards issues as the ability to operate in real-time, handling of huge quantities of information and procedures to handle temporal and non-monotonic reasoning. Whenever there is a need for those specific functionalities, a new inference engine must be done from scratch. This paper presents the inference engine developed for SPARSE, an expert system for intelligent alarm processing and power restoration aid which is running on-line in a Portuguese transmission control center.