Ensuring exam integrity with AI-assisted remote proctoring
Overview
With the world swiftly transitioning into the digital era, online exams have become an essential requirement across education, competitive testing and professional certifications. This shift has further heightened the demand for ensuring exam integrity. A leading US-based publication identified that the heavily manual and resource-intensive traditional exam proctoring models were not able to scale to meet the growing need for secure, reliable and accessible online exams.
HCLTech’s AI Engineering team developed an AI-assisted remote proctoring system capable of analyzing live webcam feeds in real-time, even from low-configuration devices with basic webcams (<1 megapixel). This innovation enabled scalable monitoring, real-time detection of prohibited items and suspicious behavior and centralized oversight across thousands of concurrent examinees.
This case study explores how our collaboration transformed the exam monitoring process from reducing proctor-to-examinee ratio and operational costs, to improving integrity, responsiveness and trust in high-stakes digital assessments.
The challenge
Our client, a multinational education provider, runs large, time‑bound exam windows with candidates worldwide. Scaling fairness and process consistency presented specific hurdles.
While manual supervision was resource‑heavy and inconsistent across programs, the candidate devices often had basic webcams and older hardware, thereby limiting detection accuracy.
The malpractice risks, including usage of prohibited items (phones, smartwatches, calculators, notes) or concerning behaviors (multiple faces, gaze shifts, tilted faces, or empty frames) demanded precise and low‑latency monitoring.
Inefficient detection systems produced frequent false positives, distracting proctors and inflating operational effort, therefore creating inefficiency.
The demand surge required the system to handle thousands of concurrent streams in a reliable manner.
To meet these realities, the client needed an AI‑enabled proctoring system that was accurate, scalable and cost‑efficient, with clear audit trails and minimal friction for examinees.
Our solution
HCLTech collaborated with the client to develop an AI‑assisted proctoring system that enhanced the Client Proctoring System (CPS) interaction. The system is supported by a plug-and-play microservices-based architecture that caters to future scaling demands.
The solution was built with custom AI models designed to capture features at multiple scales using a pyramid structure. This reduced scaling and positional errors, making the system effective even with lower-resolution video feeds. Residual layers extended the model’s field of view, ensuring better detection of objects and behaviors in real exam conditions. A single-shot regression method was also used to draw bounding boxes and assign class labels, allowing the network to identify prohibited items and activities in real time through a single, efficient pipeline.
key highlights
- Live monitoring of exam sessions with <2 second response time for malpractice detection
- Automatic detection and alerts for prohibited objects, such as phones, smartwatches, calculators, books and writing instruments
- Eye gaze tracking, head tilt monitoring and single/multi-face detection to flag suspicious activity
- Verified input stream quality for AI-worthiness before processing
- Automatic start-end event markers for auditability and post-exam reviews
Impact delivered
Smarter and secure online exams
- 1:115 proctor-to-examinee ratio, improved from 1:10, significantly reducing human dependency
- False positive rate below 10% to keep proctors focused on genuine risks instead of being bothered by false alerts
- Real-time detection of prohibited objects and unauthorized persons for greater exam credibility
- Stream‑quality checks stabilized accuracy on low‑resolution webcams and varied environments
Cost‑efficient operations at scale
- <$0.50 per stream analytics cost, making large-scale proctoring affordable
- Central monitoring of multiple examinees worldwide with minimum infrastructure to support remote proctoring
- Reduced manpower and operational overhead, freeing resources for other critical academic functions
Built for expansion
- 10,000 webcam video streams managed simultaneously to handle burst load during seasonal peaks
- Completely configurable AI system to support efficient operations and seamless scalability
As a trusted engineering partner, we helped our client enable secure, real-time and cost-efficient monitoring of online examinations, setting a new benchmark for fairness and reliability in remote assessments.
