Building a skill-assessment engine for hiring: timed tests, anti-cheat, and category-based matching

I have been developing HireLyf, a hiring platform, and I am eager to explain the workings of the core assessment engine, as it presented the most challenging aspect to perfect. The primary concern was to provide candidates with a fair, difficult-to-cheat skill test that accurately predicts job readiness, all without relying on a comprehensive proctoring service.

My solution involves a 40-question, 20-minute assessment divided into ten-question sections, dynamically pulled from a question bank covering SQL, coding, data structures, Python, React, and more. Candidates select their own test categories based on their listed skills, with English and aptitude always required. To prevent cheating, the platform incorporates an anti-cheat layer utilizing client-side signals such as tab-switch detection, window-blur tracking, copy/paste blocking, and cursor-leave detection.

These signals are amalgamated into a flag threshold, rather than triggering a hard block, as individual signals alone result in excessive false positives.

The scoring system remains out of 100, with 25 points awarded per section, irrespective of the number of questions presented. This ensures that the math remains consistent regardless of the test duration. The technology stack includes Next.js 14, MongoDB via Prisma, and is deployed on Vercel. I am willing to provide further details on any aspect of this process, particularly the anti-cheat calibration, which required multiple iterations to strike a balance between detecting deceitful behavior and avoiding false positives on the part of honest candidates.

For those interested in the full platform, it is available for early access at https://www.hirelyf.com.