interview.cthruai.com

Technical interviews for the age of AI.

Stop testing whether candidates can memorize algorithms. Test how they actually work now: prompting AI, catching its mistakes, and applying real engineering judgment.

No account required to create or attempt a problem. Sign in only to keep your history.

Have a problem link?

CodingSystem designSpec writingCode reviewBrownfield refactor

AI-native problems

Candidates use a built-in AI assistant to generate code, designs, and specs. You evaluate how well they prompt, iterate, and catch hallucinations.

Tunable generation

Dial in difficulty, size, language, and inject anti-patterns: memory leaks, race conditions, deadlocks, N+1s, security flaws, misleading comments, and hallucination bait.

Brownfield & review

Drop candidates into legacy code to grok and refactor, or hand them a diff to review with inline, line-anchored comments.

Real execution

Build and run code in a sandboxed multi-language environment. Output streams back to a clean editor UI.

Rigorous evaluation

AI-assisted rubric scoring of spec quality, review-comment validity, correctness, and AI-usage skill, with full human override.

Session replay

Every prompt, AI response, edit, run, and comment is captured on a timeline so you can replay exactly how a candidate worked.