Scalable Automated Proctoring: Integrating Browser-Native Artificial Intelligence and Continuous Trust Evaluation

Abstract

Abstract—The ability of most insitutions to make an agreement
swiftly to remote education highlighted numerous integrity issues,
thus established the need for an application-wise authentic
proctoring strategy. Nonetheless, present automated solution for
proctoring is either based upon sophisticated deep convolutional
network (such a deep YOLO network) time-consuming requires
great deal of computer power or constantly intrusive online
desktop application to integrate the knowledge from students not
being located together with test location. Many of these approach
disqualify the students utilizing common computer parts they
were in all probability to use to even go on the web with, com
promise the citizens of those pupils, two the ability to meddle with
evidence acquired on a far more powerful proctoring solution.
We propose a stream-lined, wholly virtual multi-modal AI based
Proctoring system integrated to a Django platform (PostgreSQL
server for data storage) to get over those limitations. In this
project, we demonstrated a system of make sure the proctoring
software unwilling run while offline onto a pupil computer. This
project we used the enormous computer-processing and memory
learning pipeline is similar to repacement by existing proctoring
systems with a lean angular multiplier modality spatially-tracer
system using Face–based facial recognitions of pupil to do Gaze
Traccing (em the result straight away) for objects detection
during image minutely track for obiquitous objects on pupil
view(s). Fourth, using all of those background sound analysis
plus ongoing browser app up to a modern ”Trust Value”. The
trust value will evaluate the behavior of student through out the
whole duration exam. This propose to features a way to design
a proctoring system that utilizes minimal amount of bandwidth
and computational resource, can be expanded to accomodate
many students taking a test simulteneously and comport to
accommodate with the student wellbeing- rather than likely make
any of these metrics deteriorate.
Index Terms—Remote Proctoring, Uncertainty Estimation,
Multi-Modal Fusion, YOLOv11, LSTM, Edge Computing.