Practise AI-assisted coding the way real interviews test it.
Work in a real codebase with an AI agent, catch what it gets wrong, verify with hidden tests, and get scored on your judgment.
+ user["active"] = False
Work the way you do on the job
One agent, one unfamiliar repo, and no script for how to use them.
An agent you talk to
Ask it to explain code, chase a failure, or build the feature, in your own words. It reads files, edits them and runs your tests, like the coding assistants you use at work.
A repo you've never seen
A real multi-file codebase in a VS Code-style editor, with a terminal and tests. You decide how to get oriented, and hints are there only if you ask and spend one.
Your judgment is what's scored
Every edit the agent makes is yours to inspect and undo. We score how you directed it, what you checked yourself, and whether the final code actually works.
Five kinds of real work
In Python, TypeScript, Java and C++, across APIs, SQL, security, concurrency and performance. Not algorithm puzzles: the actual shapes of AI-assisted engineering.
Debug
A real multi-file repo with a genuine bug and a failing hidden test. Find it, fix it, prove it.
Build
A scaffold and a spec with room for interpretation. Ship something that passes the acceptance tests.
Extend
A working app and a new feature request. Add it without breaking what already works.
Review
An AI-generated pull request with planted issues. Catch what a careless reviewer would miss.
Reason
Design and systems questions on caching, indexes, retries and queues, answered and graded without writing code.
More than a single problem
Practice, measure, and prepare for the interview itself, in one place.
Timed assessments and mock interviews
Adaptive timed tests and hard-stop mock interviews with a hiring-style verdict.
Contests
Scheduled rounds with a leaderboard, for when practice should feel like pressure.
Communication practice
STAR answers, project walk-throughs and HR questions, with feedback on what you wrote or said.
A plan for your target role
Readiness by area, a countdown to your interview date, and a study path that follows your weak spots.
Scored on judgment, not vibes
Every report combines real hidden-test results with a rubric built from your actual session, plus an AI coach narrative pointing at the exact moments that mattered.
Real hidden tests run against your final code, not a self-report.
Did the agent's changes get tested, or did you take its word for it?
Checked the agent's work yourself, and undid or fixed what was wrong.
How clearly and efficiently you steered the AI toward a real solution.
Lasting changes per message: fewer wasted round trips for the same outcome.
No network access, no shared state between sessions, no way for an AI-generated command to reach anything but your own throwaway container.
Ready to see how you actually direct an AI?
Debug, build, extend, review and reason through problems in Python, TypeScript, Java and C++.