Despite many potential applications of Building Information Modeling (BIM) in operation and maintenance of existing buildings and infrastructure, digital as-built models are rarely available. However, the manual evaluation and reconstruction of as-built models from available data as images, point clouds, 2D plans, and text documents is time-consuming and costly. A key aspect of obtaining a digital twin is fully automated data acquisition and processing, which raises the need for intensive research in the field of automated generation of digital as-built models. Our team develops various artificial intelligence (AI) methods to generate a virtual replica of construction projects. For example, computer vision methods are used to capture the geometry of existing structures. In this regard, point clouds from LiDAR scans and images are processed to capture the as-is geometry. In addition, 2D construction plans and CAD files are used to extract geometries and information to merge them into 3D models. Natural language processing techniques allow for the analysis of construction documentation, which also provides important information on the current condition of buildings. AI models can learn to interpret this data and assign it to individual building components. This results in a digital twin with higher information content.