Over the last few decades, building development has been recorded using hand-made blueprints before CAD tools appeared and later with digital building plans. As a consequence, there is a large amount of information in millions of assets that are hard to process because of their analog nature. Since adopting the Building Information Model (BIM) approach, any new building plan can be subject to sophisticated validations and analysis. However, legacy analog plans can not profit from sophisticated BIM analysis, and it is not feasible to manually generate BIM representations at low cost. There is a demand for BIM models of existing buildings that are feasible to be integrated into a workflow for building energy retrofitting. This paper presents a novel approach to generating BIM Models based on artificial intelligence algorithms by parsing architectural and structural drawings. To identify elements from blueprints and generate the model, we first trained the Mask R-CNN framework with our dataset of 9 concrete buildings composed of architectural and structural blueprints. The outcome of the process is a BIM model corresponding to one of the multi-story buildings using the Industry Foundation Classes (IFC) format. Building development has been recorded using hand-made blueprints before CAD tools appeared and later with digital building plans.