Agentic AI for Human Resources: LLM-Driven Candidate Assessment

A research team from institutions including Mohamed bin Zayed University of AI has published a paper titled 'Agentic AI for Human Resources: LLM-Driven Candidate Assessment' at EACL 2026. The work introduces a modular, interpretable framework that leverages Large Language Models (LLMs) to automate and enhance the recruitment process. Unlike traditional Applicant Tracking Systems (ATS) that rely on simple keyword matching, this system employs a multi-agent architecture. Each agent is designed with role-specific, LLM-generated rubrics to perform fine-grained, criteria-driven evaluations by integrating diverse data sources like job descriptions, CVs, interview transcripts, and HR feedback. The output is a detailed, transparent, and auditable assessment report suitable for real-world hiring workflows.

Beyond rubric-based analysis, the paper's core technical contribution is a novel ranking mechanism called the 'LLM-Driven Active Listwise Tournament.' This approach addresses the noise and inconsistency of traditional pairwise comparisons or independent scoring. Instead, the LLM is tasked with ranking small, manageable subsets of candidates (mini-tournaments). These listwise permutations are then aggregated into a globally coherent ranking using a statistical Plackett-Luce model. An active-learning loop selects the most informative subsets for the LLM to evaluate, making the ranking process highly sample-efficient. This methodology, adapted from financial asset ranking, provides a principled and interpretable way to handle large-scale candidate ranking in talent acquisition, moving beyond black-box scoring.