Review take-home codein 60 seconds.

Drop an assignment brief and a GitHub URL. CodeVerdict maps every requirement to code, detects AI shortcuts, and generates interview questions.

Runs whatever they builtAuto-detected and booted in an isolated VM.
React
Next.js
Vite
TypeScript
Node.js
Express
NestJS
FastAPI
Django
Flask
Go
Rust
Postgres
MongoDB
Redis
MySQL
Docker
Docker Compose
Watch

See a verdict in real time.

codeverdict.io · live demo · 60 seconds
Demo video coming soon

↑ 60-second walkthrough · sound off · loops

How it works

From brief to verdict in three steps.

1

Drop the brief, paste the URL

Upload your brief. Paste a GitHub URL. That's the setup.

2

Read, run, score

Clone, install, boot, check — in an isolated VM.

3

Read the verdict

Scored report, file:line evidence, interview questions ready.

Features

Everything you need to trust the verdict.

Requirement mapping

Every brief line → file:line that satisfies it.

Sandboxed execution

Boots on ephemeral VMs. Broken repos surface fast.

AI shortcut detection

Perplexity + entropy + commit history. With evidence.

Grounded interview Qs

5–10 questions per submission, tied to actual code.

Side-by-side compare

Pin four candidates. Pick the strongest in 30 seconds.

Custom scoring weights

Weight every signal your way. Verdicts update instantly.

Evidence

Proof, not opinion.

Every check anchored to file:line. Every AI flag pointed at the suspect block. Every score a sum of signals you can audit.

Requirement check · live4 of 5 met
Implement JWT auth with refresh tokens
src/middleware/auth.ts : 18
Rate-limit /api/* to 60 req/min per IP
src/middleware/rate-limit.ts : 23
Persist sessions to Postgres, not memory
src/db/sessions.repo.ts : 7
WebSocket reconnect with backoff
src/lib/socket.ts : 47 — partial (no jitter)
Tests covering the auth middleware
No test file references middleware/auth.ts
FAQ

Questions, answered.

How long does a verdict actually take?
Most submissions complete in 60 seconds. The agent clones the repo, installs dependencies, boots the app in an E2B sandbox, runs the brief's checks, and writes the report — in parallel. Larger repos with heavy install steps can take up to two minutes.
GitHub repos, zip files, or both?
Both. Paste GitHub URLs (public or private with a token), or upload a zip. CodeVerdict treats them identically: clone, install, run, evaluate.
Which stacks are supported?
Node / TypeScript, Python, Go, plus any Docker / docker-compose project. The agent auto-detects the stack from project hints (package.json, requirements.txt, go.mod, Dockerfile) and picks the right setup path. Frontend SPAs, full-stack apps, and REST APIs all work.
How does AI-code detection work?
Three signals combined: token-level perplexity (LLM-written code has unusually uniform probability distributions), naming and structural entropy, and commit-history analysis (sudden monolithic commits with no incremental work). Output is a 0–100 AI-share score with file:line attribution — no black box.
Is candidate code safe to run?
Yes. Every repository runs inside a single-use E2B virtual machine that is destroyed immediately after the report is generated. Candidate code never touches your machine or CodeVerdict's production servers.
Can I customise scoring?
Yes. Weight requirements, quality, tests, security, and AI-share independently. The verdict (Strong hire / Hire / Maybe / No hire) is computed from your weightings and updates instantly across every past submission.

Stop reading code.
Start hiring engineers.

Free during beta. Join hundreds of technical teams using CodeVerdict to automate their take-home review.