Hi HN! I’m not a software engineer by trade. My background is in behavioral science and human-computer interaction, but I wanted to solve a specific problem I found when applying for jobs which is the logical test. They are booooring, easy to cheat on if you really want to and provides very limited information about people. I built this prototype from scratch sing AI to bring my idea to life. So now I've been working on this side project for a Cognitive Adaption Engine (CAT/KAT in swedish) for about three weeks. Many pre-employment screening tests i've done use a static item bank and are very liberal with the time allowed on each question, making it easy for Vision LLMs or bots to beat. I wanted to see if I could build a fast, client-side engine (JS/SVG) that combats this problem by measuring certain biometric cognitive signatures like "switch cost" (how long it takes for your brain to re-adjust when logical rules change) and "post error-slowing". To go beyond the simple grading of correct/incorrect, this test dynamically gives you waves of patterns/challenges through a local item bank based on your millisecond response times. If you use an LLM, the constant latencies gets punished with a lower score. If they use a script, the complete lack of natural human variance will be picked up on. I am currently trying to calibrate a baseline and need high-aptitude users to stress-test it and to provide more data. It also makes it very fast to complete.

Everything runs 100% locally in your browser and since I did use AI, I would love your feedback on how the architecture holds up and the technical approach to anti-cheating or anything that comes to mind on this topic.

(Optional) When you finish the test, you can press Ctrl + Shift + X on the results page which unlocks a hidden admin console. There you can see more info on your performance and even download a HR-pdf that translate your data into three tailored interview questions. Unfortunately the admin page is in Swedish, but I believe you will get the idea.