Urban environments significantly influence residents’ health, well-being, and community engagement. However, current methods for assessing walkability often lack efficiency and accuracy, especially given the challenges posed by urban expansion and environmental shifts. This study presents an automated framework that integrates deep learning techniques with geographic data to evaluate walkability and recommend optimal pedestrian paths. We explore two approaches: context-based learning using GPT-4o mini and similarity-based learning with Contrastive Language-Image Pretraining (CLIP). By analyzing Google Street View images, we generate walkability scores for specific locations and employ a ranking system that combines these scores with geographic features, such as elevation and slope, to identify the most accessible paths. Our comprehensive evaluation across several Seattle neighborhoods, compared with human perception data from surveys, demonstrates the effectiveness of our approach. This scalable framework enhances walkability assessments and supports the creation of safer, more inclusive urban environments.