Extending the existing building stocks is a principal task to meet the enormous demand for inner-city living, minimize the need for additional transport infrastructure, and reduce environmental pollution and land consumption. The Cluster of Excellence Integrative Computational Design and Construction for Architecture (IntCDC) at the University of Stuttgart focuses on this topic through a co-design approach for integrating computational design, engineering, and robotic construction. To enable construction in existing buildings, a reliable representation of the as-built geometry of the building is needed. However, this data is often not available, but the design plans are available, which may not always match the current state of the building, or there may have been deviations already during construction. This paper brings forward a novel approach for the geometry representation, taking advantage of different geometry sources and their stochastic models, including floor plans, Terrestrial Laser Scanning (TLS) and Simultaneous Localization and Mapping (SLAM). Initially, the as-planned geometry was modeled digitally based on the floor plans and considering the stochastic information. In this contribution, the Building and Habitats object Model (BHoM) was utilized instead of data schemas like Industry Foundation Classes (IFC) since it fulfills the specific co-designing requirements. Secondly, the generated TLS and SLAM point clouds were processed, and stochastic models were attached. Finally, to generate a comprehensive geometry representation, the segmented geometric primitives from different sources were fused based on the respective approximated stochastic models. The results were integrated into the BHoM for expansion purposes.