Constructing a building is a complex and decentralized task, often resulting in deviations from the original design due to cumulative human errors. In the United States, the responsibility for overseeing and identifying these construction deviations mainly falls on the site supervisor and general contractors, who visually and manually inspect the construction’s progress. This method is time-consuming, expensive, and offers limited productivity and accuracy. Additionally, it often fails to document minor defects that can escalate into significant issues. Over the past decade, quadruped robots like Boston Dynamics’ Spot, ANYbotics’ ANYmal, and Unitree’s Go1 have made it possible to collect accurate, autonomous data in the cluttered and uneven terrains typical of construction sites. Their high payload capacity, powerful onboard computing, and ability to navigate complex environments give them a clear advantage over robotic platforms, such as drones or wheeled robots, which often face limitations related to stability, modularity, or regulatory restrictions. Despite these capabilities, much of the work to process the large, multi-sensor datasets and inpoint as-built deviations from the intended design largely remains a manual, post-processing endeavor. This thesis aims to address the problem of manual post-processing required to identify as-built deviations. To generate an as-built construction environment from a scanned dataset, multi-sensor Simultaneous Localization and Mapping (SLAM) is necessary. SLAM approaches inherently contain scene reconstruction errors. This work evaluates the reconstruction accuracy of a popular SLAM library, particularly concerning construction site component tolerances and sensor coverage limitations, to determine the practical limitations of using a quadruped robot to measure construction deviations. The research presented here is a structured prototype confined to simulations conducted within NVIDIA IsaacSim (Version 2023.1.1), utilizing the NVIDIA Orbit modular framework. The initial stage involved creating a realistic construction site asset and simulating an ANYmal C quadruped robot on this site. The robot is user-navigated, and the simulated sensor data is collected for post-processing through LiDAR SLAM methods to generate as-built scan reconstructions. This reconstruction is then assessed for deviations and displayed in a visually digestible manner, allowing users to see which components deviate from the intended design.