This thesis designs, evaluates, and deploys fairness-aware artificial intelligence (AI) systems for recidivism prediction and other high-stakes domains. It addresses persistent gaps in the literature by integrating fairness interventions across the AI lifecycle, improving fairness–accuracy trade-offs, and developing methods that expose intersectional harms. The research advances multi-phase fairness pipelines that integrate pre-, in-, and post-processing techniques, supported by optimizationinformed evaluation and intersectional auditing frameworks. Grounded in trustworthy AI and socio-technical governance principles, the study aligns fairness interventions with ethical and procedural standards for responsible AI in criminal justice and beyond. Empirical validation uses four datasets: COMPAS, RisCanvi (a curated dataset), Adult Income, and an Irish health insurance dataset. Standardized protocols and fairness metrics, Statistical Parity Difference, Disparate Impact, Equal Opportunity Difference, and Predictive Equality Difference, guide analysis and model evaluation. The thesis extends the AIF360 toolkit through enhanced algorithms (DIR+,AD+) capable of handling multi-valued protected attributes and improving training stability. Results show that integrated pipelines outperform isolated methods, achieving consistent fairness gains with limited accuracy reduction. Fairness-aware oversampling mitigates subgroup imbalance but requires careful validation to avoid synthetic bias. Intersectional auditing reveals disparities masked in aggregate measures and highlights the contextual nature of fairness. Conceptually, the thesis contributes the Justice-Aware AI Fairness (JAAF) framework, linking technical fairness methods with intersectional and governance-based accountability. Overall, it delivers integrated fairness pipelines, optimization-based evaluation tools, and guidelines for fair, transparent, and trustworthy AI systems in high-stakes decision-making.