Tired of candidates using AI assisted software?
Share one window. The question is a picture an AI assistant cannot read, and the plain text beside it is a decoy written for the assistant to answer.
One shared window. The answer only a person can read.
The interviewer shares a single window. The question is an image; the decoys beside it are the bait. A parroted answer gives the assistant away.
Share one window
Screen-share the present view in Zoom or Teams. Your private control window, with the real question and notes, stays off the call.
The question is a picture
Each question renders as an image built to defeat OCR and vision models. A person reads it at a glance; a screenshot tool cannot.
Decoys catch the tool
Plain-text decoy questions sit beside the image. An answer that follows the decoy, not the picture, exposes the assistant.
Built so a vision model reads the wrong thing.
Every account gets its own designs, every slide carries decoys, and every design is tested against vision models nightly.
A style of your own
Every account is assigned its own captcha designs. Nobody can sign up, collect the catalogue and train an assistant against it.
Decoys on every slide
Every slide carries near-miss questions in plain text, and your private window lists exactly which decoys the candidate was shown.
OCR is broken
Hatching, occlusion, displacement and per-frame warping. A person reads the question. A character recogniser reads fragments.
Canary words
Each decoy block ends with a planted word to use in the answer. A candidate reading the picture never sees it. A tool follows the instruction.
Pointer and background noise
The shared pointer is a shape that also drifts down the page in colours barely visible to the eye, beside decoy text, so a tool watching the screen is fed noise.
Read by vision models nightly
Every design is rendered at screen-share quality and read by the vision models a candidate could use. Any design read three runs in a row is retired and replaced.