Xylazine is a veterinary sedative that has emerged as a common adulterant of illicitly-manufactured fentanyl. Xylazine contributes to ulcer development, with variable characterization. Advanced xylazine-associated ulcers present severe tissue necrosis and are managed differently from alternative wounds. Clinical care of potential xylazine-associated ulcers is difficult: (1) Xylazine Exposure Detection: Discrimination of xylazine-associated ulcers from alternative wounds is difficult, given that xylazine testing is rarely available. (2) Superinfection Assessment: Because of their necrotic presentation, it is difficult for clinicians to assess wound superinfection likelihood and appropriately suggest systemic antibiotic usage. In this study, we aimed to (1) define the performance characteristics of a publicly-trained deep learning tool for accurate classification of xylazine-associated wounds from provider-obtained clinical images, and (2) develop, train, and clinically validate a deep learning tool to accurately classify wounds with bacterial superinfection requiring systemic antibiotics. Methods associated with this study that build upon previous work include the development and clinical validation of an AI-based pipeline for clinical assessment of potential xylazine wounds that spans both xylazine exposure determination and superinfection assessment for improving antibiotic stewardship. Using performance characteristics on a publicly-sourced dataset used to train different Model 1 network architectures, we identified an architecture (EfficientNetB0) that differentially learned features unique to xylazine-associated ulcers and exhibits high performance and specificity. We identified and obtained wound imaging from N=450 and N=150 Penn Medicine patients (from a starting cohort of UPHS xylazine-labelled visits) to complete Aims 1 and 2, respectively, after applying a set of inclusion and exclusion criteria designed to isolate a rigorous clinical dataset. Our work presented here will support the creation of a comprehensive registry of xylazine-associated wounds, allowing for evaluation of utilization of healthcare interventions, and longitudinal outcomes of xylazine-associated wounds with or without bacterial superinfection. Furthermore, the proposed deep learning tool pipeline has the potential for integration within an EMR, providing a scalable AI-based clinical model for xylazine wound assessment that other healthcare systems can adopt as the xylazine crisis propagates to other regions.