Intervivy: A Multimodal Voice AI Platform for Intelligent, Accessible, and Ethical Remote Recruitment

This paper presented Intervivy, a modular, multimodal AI platform for intelligent remote recruitment that addresses the three dominant failure modes in contemporary AI interview systems: infrastructure fragility, inadequate malpractice detection, and evaluative reductionism. Through the integration of WebRTC and Supabase, the platform sustains session continuity across adverse network conditions with sub-4-second recovery times. The Multimodal Malpractice Detection System, combining MediaPipe Face Mesh gaze analysis, YOLOv8-Nano object detection, and a Heuristic Fusion Engine, achieves 94.5% detection precision with fewer than 5% false alarms satisfying demanding real-world proctoring requirements. The Whisper-Gemini evaluation pipeline replaces binary hiring verdicts with rubric-grounded, skill-disaggregated reports at 93% alignment with expert human assessments. Taken together, these contributions establish Intervivy as a reproducible and ethically grounded architecture for AI-assisted recruitment. Future work will pursue four directions: integration of real-world candidate datasets across demographic groups to benchmark fairness and bias; A/B testing of AI versus human evaluation consistency across diverse job roles; extension to multi-language interview support for cross-regional deployment; and incorporation of prosodic and paralinguistic features from the audio stream to enrich communication dimension scoring. The platform's modular microservice design positions it well for incremental improvement and cloud-scale deployment without architectural rearchitecting.

[1] L. Chen and H. Acosta, "The Rise of the Asynchronous Video Interview: A Study on Fairness, Candidate Perceptions, and Predictive Validity," Journal of Applied Psychology, 2024. [2] R. Sharma, S. Gupta, and K. Rao, "Evaluating the Efficacy of Large Language Models for Automated Skill Assessment in Technical Interviews," in Proc. ACM CHI Conference, 2023. [3] J. Adebayo and K. Williams, "Mitigating Gender and Racial Bias in Automated Video Interview Scoring," IEEE Transactions on Affective Computing, 2023. [4] M. Vazquez and A. Singh, "A Multi-Modal Approach to Predicting Job Performance from Video Interviews Using Audio, Text, and Visual Cues," in Proc. ICMI Conference, 2022. [5] M. Ibrahim and J. Deng, "A Comprehensive Survey of Online Proctoring Systems: Architectures, Techniques, and Challenges," ACM Computing Surveys, 2024. [6] F. O'Reilly, R. Thomas, and H. Zhang, "A Heuristic-Based Anomaly Detection Engine for Identifying Suspicious Behavior in Remote Assessments," Journal of Educational Data Mining, 2022. [7] V. Patel and J. Kim, "Gaze-Based Attention Tracking for Online Proctoring Using MediaPipe Face Mesh," in Proc. IEEE/CVF CVPR, 2024. [8] A. Petrov and S. Nielsen, "Retell AI and Conversational Voice Agents: A Low-Latency Architecture for Turn-by-Turn Dialogue," in Proc. EMNLP Conference, 2023. [9] J. Martinez and B. Chen, "Adaptive Bitrate Algorithms for Real-Time Video Communication over Unstable Networks," IEEE Transactions on Mobile Computing, 2022. [10] T. Schmidt and C. Wagner, "Resilient Web Applications: A Framework for Session State Management and Resumption after Network Disconnections," in Proc. ACM Web Engineering, 2023.