Recent innovations in LiDAR sensors and machine learning have significantly improved our ability to capture and label powerline infrastructure in 3D. By leveraging high-density LiDAR data acquired by drones, small yet critical elements such as insulators, attachment points, conductor and guard cables can be detected and classified with better accuracy. However, simply annotating point clouds does not directly translate to ready-to-use geographic database elements, prompting a need for reliable vectorization strategies.In this work, we propose a workflow that combines clustering, filtering, and geometric modeling to convert semantically segmented point clouds into structured vector objects.1 We validate our approach in both manually annotated datasets and automatically segmented data via a deep learning method (Superpoint Transformer), achieving consistent reconstruction of pylons, cables, and auxiliary components across two distinct test sites.