Mobile laser scanning technology enables the rapid acquisition of highly detailed 3D representations of road environments in a single survey pass while travelling at normal road speeds. The efficiency of this data collection process has transformed surveying workflows across many engineering projects. Despite these advances, feature extraction, particularly the generation of CAD- and GIS-compatible outputs, remains a significant challenge. This is especially true for linear roadway assets such as curbs and lane markings. Extracting these features from mobile LiDAR datasets is an extremely tedious and time-consuming process. Although semi-automated tools have helped, they still require substantial user intervention and many hours of manual drafting. To address these limitations, this presentation introduces Reflektar’s novel AI framework for the fully automated, network-scale extraction of curbs and pavement markings from mobile LiDAR point clouds. The framework combines supervised deep learning for feature detection with an unsupervised machine learning pipeline that vectorizes semantic attributes based on the detection results. The outputs include CAD- and GIS-compatible representations of roadway features such as top and bottom of curb, , lane lines, dashed markings, arrows, and other pavement marking symbols. By eliminating the need for manual interaction, the proposed framework enables scalable deployment across hundreds of kilometers of roadway. The presentation describes the proposed framework and shares the results of citywide deployment of the novel Reflektar framework. The presentation also discusses challenging boundary cases. Results from testing across diverse environments and multiple mobile mapping systems are also presented.