Ethical Bias in AI Datasets: A Data Science Audit Framework for Fairness and Transparency (DS-Audit v1.0)

As AI systems are increasingly deployed in hiring, finance, and governance, ethical bias embedded in training datasets poses a critical risk to fairness and transparency. This technical report presents DS-Audit v1.0, a five-phase Data Science Audit Framework for systematic evaluation and mitigation of dataset bias. The framework is designed for independent researchers, startups, and academic institutions to perform responsible AI auditing without enterprise infrastructure. A case study in the Indian IT hiring context demonstrates practical applicability, where Disparate Impact improved from 0.41 to 0.88 after mitigation.