The ability to forecast energy demand is an essential part of optimizing the energy supply chain as it enables network operators to meet a wide range of security and planning objectives. However, forecasting studies are often hampered by a lack of readily available data. In Colombia, long-term forecasts of at least one year have traditionally been made with various statistical models. Recent developments, however, have made it possible to build a comprehensive multi-year database thanks to numerous AMI sensors. Based on this extensive, hourly-resolved database, this paper examines short-term predictive modelling of energy demand levels. The data is meticulously processed to create daily time series that are then used in predictive models. Several machine learning techniques for time series forecasting within a one-week short-term forecasting framework are thoroughly explored. The findings highlight the potential of classical kernel-based machine learning models, such as support vector machines and Gaussian processes, in modelling and forecasting short-term energy demand.