The increasing deployment of machine learning (ML) systems in high-stakes domains such as healthcare, finance, criminal justice, and employment has significantly amplified concerns around bias, fairness, and accountability, as these systems increasingly influence decisions that affect people's lives, opportunities, and rights. While ML models are often promoted as more efficient, consistent, and objective than human decision-making, they are deeply shaped by the data they are trained on, the objectives they are optimized for, and the institutional contexts in which they are deployed, meaning they can inherit, reproduce, and even amplify existing societal biases and power asymmetries. At the same time, the opacity of many modern ML models, particularly deep learning systems, has raised challenges for interpretability, transparency, and trust, making it difficult for stakeholders to understand, contest, or audit automated decisions. In response to these challenges, this article proposes a Responsible AI Framework that integrates three interconnected pillars: (1) formal fairness definitions and quantitative metrics to systematically identify and measure bias, (2) documentation-based accountability through structured artifacts such as datasheets for datasets and model cards for models to enhance transparency and reproducibility, and (3) governance and auditing mechanisms at organizational and policy levels to ensure ethical alignment, oversight, and compliance. Drawing on key studies published between 2000 and 2021, and informed by three foundational diagrams fairness trade-offs, model cards, and datasheets this article argues that responsible AI cannot be achieved through technical solutions alone but instead requires a socio-technical approach that meaningfully combines technical rigor with institutional accountability, stakeholder participation, and ethical governance.