A Survey of AI-Powered Multi-Agent Interview Platforms for Candidate Preparation and Intelligent Recruitment

This survey has provided a systematic and critical review of the current landscape of AI-powered interview systems, examining the state of the art across three foundational technological streams—multimodal behavioral analytics, large language model applications, and multi-agent architectures—and synthesizing these perspectives within the broader context of intelligent, AI-driven interview ecosystems.   The literature consistently demonstrates that multimodal approaches substantially outperform unimodal baselines in candidate assessment, though the performance advantage is most pronounced for objective behavioral metrics and diminishes for complex psychological constructs. LLM fine-tuning for interview-specific domains produces dramatic improvements in question quality and contextual appropriateness, and multi-agent architectures provide the structural flexibility needed to manage end-to-end interview complexity. When integrated, these three streams offer a credible path toward interview platforms that are simultaneously more efficient, more consistent, and more informative than traditional human-only processes.   The field, however, faces substantial and interconnected challenges. The technical difficulty of real-time multimodal fusion at scale, the persistent problem of algorithmic bias and limited model generalizability, and the ethical imperatives of transparency, explainability, and data protection collectively require sustained interdisciplinary effort. Progress on these fronts depends on collaboration between AI researchers, HR professionals, organizational psychologists, legal scholars, and the candidates whom these systems are designed to serve. The future of AI-powered interview systems lies not in the wholesale replacement of human judgment, but in the thoughtful augmentation of it—creating evaluation frameworks that harness the precision and consistency of artificial intelligence while preserving the empathy, contextual sensitivity, and ethical accountability that remain distinctively human contributions.  

[1] A. I. Uffcutt and D. J. Woehr, “Further Analysis of Employment Interview Validity: A Quantitative Evaluation of Interviewer-Related Structuring Methods,” Journal of Organizational Behavior, vol. 20, no. 4, pp. 549–560, 1999. [2] A. Agrawal, R. A. George, S. S. Ravi, S. Kamath, and A. K. M., “Leveraging Multimodal Behavioral Analytics for Automated Job Interview Performance Assessment and Feedback,” arXiv preprint arXiv:2006.07909, 2020. [3] X. Quan, Y. Zhang, X. Xie, and Y. Wang, “An Interview System Based on Large Language Models and Multi-Agent Interactions,” in Proc. 20th IEEE Int. Conf. on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD), 2024, pp. 1–8, doi: 10.1109/ICNC-FSKD64080.2024.10702317. [4] J. Li, Y. Wang, W. Qian, J. Hu, Z. Hu, R. Hong, and M. Wang, “Listening to the Unspoken: Exploring ‘365’ Aspects of Multimodal Interview Performance Assessment,” arXiv preprint arXiv:2507.22676, 2025. [5] H. Manasvi et al., “A Convolutional Neural Network Approach to Interview Simulation and Evaluation,” Int. Journal of Scientific & Applied Research in Technology, vol. 11, no. 5, May 2025. [6] S. Yi and T. Yamasaki, “Enhancing Online Structured Job Interviews: Personality Assessment Using Multimodal Neural Networks,” in Proc. JSAI Annual Conf., 2024. [7] M. T. Younes, O. Walid, K. Shaban, A. Hamdi, and M. Hassan, “Augmented Fine-Tuned LLMs for Enhanced Recruitment Automation,” arXiv preprint arXiv:2509.06196, 2024. [8] A. Ansari et al., “Evaluating Speech-to-Text × LLM × Text-to-Speech Combinations for AI Interview Systems,” arXiv preprint arXiv:2507.16835, 2024. [9] “Multimodal Hierarchical Attention Neural Network for Recruitment,” Int. Journal of Advanced Research in Science and Technology, vol. 13, no. 7, 2024. [10] L. S. Nguyen, A. Marcos-Ramiro, M. M. Romera, and D. Gatica-Perez, “Multimodal Analysis of Body Communication Cues in Employment Interviews,” in Proc. 15th ACM Int. Conf. on Multimodal Interaction (ICMI), 2013, pp. 437–444.