This report explores the use of AI and computer vision to automate condition assessment of roadside safety assets for UDOT. Leveraging street-level imagery, the project tested object detection models (YOLO11n) and vision-language models (Gemma 3, Llama 3.2) to classify defects, rate conditions, and evaluate design compliance for primary and secondary assets. Findings highlight the challenges of defect detection, the promise of advanced models, and opportunities to improve maintenance prioritization and roadway safety.