Human decisions, especially in high-stakes situations like university admissions, are susceptible to unconscious biases. While experts strive for objectivity, factors like anchoring bias and confirmation bias can creep in, leading to inconsistencies. But what if AI could help? Researchers are exploring how AI can augment human decision-making to create a fairer, more objective process. A new study introduces BGM-HAN, a hierarchical attention network enhanced with byte-pair encoding, gated residual connections, and multi-head attention. This model, combined with a Shortlist-Analyze-Recommend (SAR) workflow, mimics the human decision process. The SAR workflow first shortlists promising candidates. Then, an AI agent analyzes the shortlisted profiles, generating detailed summaries using a large language model. Finally, another agent integrates the original profile and the AI-generated analysis to make a final recommendation. Experiments with real-world university admissions data reveal promising results. The BGM-HAN model, coupled with the SAR workflow, outperforms traditional machine learning models, neural networks, and even large language models like GPT-4 in zero-shot classification. Impressively, the AI-driven system boosts both accuracy and F1-score by over 9.6% compared to human evaluators. While this research focuses on university admissions, it has broader implications for any field where objective, consistent decision-making is crucial, from hiring and loan applications to vendor selection. This work represents a significant step towards leveraging AI not just for automation, but for enhancing fairness and transparency in critical decisions.