To deliver powerful smart city environments, there is a requirement to analyse web produced data streams in close to real time so that city planners can employ up to date predictive models in both short and long term planning. Data cubes, fused from multiple sources provide a popular input to predictive models. A key component in this infrastructure is an efficient mechanism for transforming web data (XML or JSON) into multi-dimensional cubes. In our research, we have developed a framework for efficient transformation of XML data from multiple smart city services into DWARF cubes using a NoSQL storage engine. Our evaluation shows a high level of performance when compared to other approaches and thus, provides a platform for predictive models in a smart city environment.