Async

Rubric-scored

AI-reviewed

Take home assignments.

Send structured take homes. AI scores every submission against your rubric before your engineers spend a minute.

One assignment, twenty seconds.

From send to scored review — the pipeline your team actually sees.

Structure it once.

Define the task, the honest time cap, and what a strong submission looks like.

Assignment3 h expected · async
Task
Repair a flaky payment webhook handler
Make retries idempotent, keep the public interface, add tests for the failure paths.
Rubric
CorrectnessCode qualityTestsExplanation

Task and rubric defined by your team.

Four scores. Real evidence.

Every score opens to the actual work behind it.

Correctness

Does it work — including the failure paths.

Look for · failure paths covered

Code quality

Boundaries, clarity, and risk in the actual diff.

Look for · clean module boundaries

Tests

Boundary behavior proven — not promised.

Look for · edge cases exercised

Explanation

Trade-offs owned and reasoned in the write-up.

Look for · decisions explained

Where it fits.

Screen first. Assign after. Review scored work.

Screen
AI screening round
Take home
this format
Your review
humans decide

Take home questions.

The direct answers before you bring us a role.

How long should a take home assignment take?

Two to four hours is a reasonable ceiling for most roles. Longer assignments cause strong candidates to drop out. State the expected time upfront and design the task to fit it.

How does AI evaluate submissions?

Each submission is scored against the rubric you define, with written evidence attached to every score. Your team reviews pre-evaluated candidates instead of reading every raw submission.

Take home or live coding — which is better?

They measure different things. Take homes show real work quality without time pressure; live coding shows problem-solving in real time. Many teams combine them by having candidates extend their submission live.

What about AI-written submissions?

Cleared candidates walk through their own code in a short follow-up call, with AI questions built from the submission — outsourced work becomes obvious. Treat it as evidence for review, not proof of intent.

Best for.

Mid-funnel engineering hires

Writing & case-study roles

Async & multi-timezone pipelines

High-volume review teams

Real work samples — without the unreviewed pile.

Stop letting submissions pile up.

Rubric-scored reviews come back automatically — engineer time goes only to candidates who cleared the bar.