A real project, a clear brief and a set time limit. The candidate opens a working repository, reads the brief the way work actually arrives, and ships a change that a reviewer could merge.
A brief, in a real project
The task arrives the way work does: a product brief with business rules and acceptance criteria, or a ticket raised by a customer. It lands in a working project, not an empty editor. Nothing to install. The workspace runs in the browser.
AI beside the editor
A chat panel offers agent mode and plan mode. Candidates are expected to use it. The report shows what they asked, when they asked it, and how much of the output they rewrote by hand.
Visible and hidden tests
Sample tests run in the IDE while they work. A hidden suite grades the submission. Sample test names are visible. Hidden test names and inputs are never shown to the candidate.
AI interview on their own code
After they submit, about six questions are written from the diff they just shipped, mixed with questions from the role. Answers are spoken and transcribed. The interview asks how they would scale it, what they would avoid, and why they chose that design. Nothing to rehearse.
Automatic code review
Every edit, prompt, commit, and test run is recorded. Code quality is scored on structure, naming, readability, error handling, and tests the candidate wrote, with a line-level review. Nothing is graded by hand.
Set up for the role
Turn the AI assistant on or off. Choose the model, cap tokens or spend, bring your own keys, or use HirePrise. Proctoring and the built-in browser can each be switched on or off. Round length, workspace resources, and the path (DSA only, engineering only, or DSA as a gate) are yours to set.