Substance use disorders and co-occurring mental health conditions present persistent challenges for criminal justice systems, with drug courts serving as critical intervention points for nonviolent offenders. However, decision-making in drug courts is often hampered by fragmented data, inconsistent practices, and limited explainable decision support. This study develops and evaluates an AI-powered decision support system for drug courts, employing a design science methodology. The system predicts three-year recidivism risk and presents insights through an interactive dashboard. By combining SHAP-based explanations with AI-generated narratives, it translates complex machine learning outputs into actionable guidance for practitioners. Guided by task-technology fit and dual-process theories, it offers layered explanations tailored to the diverse needs of various stakeholders, supporting both rapid and deliberate decision-making. Through iterative collaboration with drug court professionals, we aligned technical capabilities with institutional constraints and ethical considerations, establishing design principles for transparent, fair, and usable decision support systems. Our effort provides three contributions: (1) a validated decision support system, (2) transferable design principles for justice and public sector contexts, and (3) empirical insights on stakeholder engagement with AI explanations. Evaluation demonstrates strong predictive accuracy, high interpretability, and enhanced user trust while facilitating more equitable treatment planning. This work supports UN Sustainable Development Goals 3 and 16 by promoting accountable decision-making that enhances judicial efficiency and treatment outcomes. • Develops an AI-driven DSS for drug treatment courts using a DSR approach. • Integrates machine learning, SHAP explanations, and Generative AI to ensure transparent and fair decision support. • Introduces a SHAP-to-GenAI pipeline that converts model outputs into personalized treatment recommendations. • Derives transferable design principles for fairness, interpretability, and usability in public-sector DSS design. • Provides empirical evidence of high predictive accuracy , interpretability, and stakeholder trust, advancing transparent, equitable justice decisions.