F*ck Leetcode.

What you can practice

Bug fix

Investigate a reported bug and fix its cause.

Feature

Build the behavior requested in an issue.

Refactor

Improve the code’s structure while preserving its behavior.

Write tests

Write tests that pass on correct code and catch planted bugs.

Code review

An unscored practice drill: read a pull request and leave line comments. Feedback checks where you comment, not what you write.

Performance

Reduce unnecessary queries or calls; checks count operations, not elapsed time.

Security

Fix vulnerabilities and handle attack inputs safely.

Coming soon

AI feedback where tests can't judge

Tests can tell whether code works. They can't tell whether a design will hold up, or whether a review would actually help a teammate. We're adding an AI reviewer for that: it will read your work against a written rubric and point at the exact lines.

• Code reviewComing soon

  It will read what your comments say, not just where they are. Did you find the real defect, explain why it matters and suggest a fix? Real issues we didn't plant count too, and noise counts against you.

• RefactorComing soon

  Once the behavior tests pass, it will judge whether your refactor reached its goal, stayed in scope and left the code easier to read.

• Write testsComing soon

  Once your tests catch every planted bug, it will judge the suite itself: whether it pins down each rule at its edges, and whether any tests catch nothing or repeat each other.

• DesignComing soon

  Open-ended problems: a business situation, a small contract the tests check, and every other decision yours. It will review your design and the DESIGN.md where you explain your tradeoffs.

Coming soon: practice for AI-assisted interviews

• Choose how much AI help you get

  No AI, syntax help only, an assistant that writes what you describe, or a full assistant. Each problem will recommend a level, like the rules of the interview you're preparing for.

• Pick your model

  Claude Haiku, Sonnet or Opus, in a chat panel beside the editor, the way most AI-assisted interviews work.

• Pull request descriptions

  Build a repo from an instructions file, then open a pull request. The AI will check every claim in your description against your code, and notice what you left out.

• Timed sessions

  Rehearse an online assessment against the clock, on a fixed problem, from first read to submit.

Tests stay in charge. They decide whether your code works; the AI only judges what tests can't. Its feedback stays unscored until it agrees with human reviewers on hand-graded examples.