Desktop Based Edge AI Proctoring System

B. Scalability and Infrastructure ChallengesCloud-based proctoring architectures consume considerable server resources due to continuous high-definition video streaming and real-time AI processing.Narayana et al. [9] emphasize that when thousands of candidates are monitored simultaneously, cloud systems experience:• System failures during peak exam periods.Heinrich [7] further highlights the lack of standardized design across commercial proctoring providers, resulting in inconsistent handling of performance bottlenecks and inefficient scaling strategies.This fragmentation creates substantial disparities in user experience, system reliability, and exam integrity across platforms. C. Bandwidth and Connectivity ConstraintsShanmugapriya et al. [10] observe that existing systems require sustained bandwidth for continuous video upload, typically between 1.5-3 Mbps for 720p/1080p streaming.In developing regions or rural environments, students frequently experience:• Frozen feeds, • Delayed alerts, • Forced exam interruptions,• System restarts or relogs.Agarwal and Sharma [25], examining emergency remote teaching, found that 41% of students globally face connectivity issues, making cloud-based proctoring inherently inequitable.The dependence on stable, high-speed internet remains one of the major barriers to fair examination conditions. D. Limitations of Single-Camera MonitoringMost commercial proctoring tools rely on a single built-in webcam, which severely restricts situational awareness and introduces blind spots.Jobby et al. [22] demonstrate that students can easily bypass monitoring by:• Looking at off-screen notes, • Keeping unauthorized devices out of camera view, • Engaging in covert collaboration.Their dual-camera offline monitoring system improved suspicious-behavior detection by 78%, confirming that single-camera setups lack holistic environmental coverage.Furthermore, Holi et al. [14] show that many cheating behaviors occur at the periphery of the camera frame-regions where singlecamera systems consistently fail to detect anomalies. E. Lack of Real-Time Explainability and FeedbackAlthough primarily a system-design issue, technical limitations also extend to the opacity of detection mechanisms.Existing systems provide:• No visual explanation for flagged events, • No real-time model interpretability, • No contextual clues for proctors, • No feedback loops for students.This results in false accusations, poor transparency, and limited appeal mechanisms.The absence of explainable AI (XAI) is a technical limitation that directly impacts fairness, trust, and usability. III.ETHICAL, PSYCHOLOGICAL, AND PRIVACY CONCERNS While technical limitations hinder the performance of online proctoring systems, the ethical, psychological, and privacy concerns documented in recent literature represent equally critical barriers to their acceptance and long-term viability.Several studies show that existing cloud-based and AI-driven proctoring tools create environments of heightened stress, perceived surveillance, and significant privacy intrusions, often disproportionate to their intended benefits.