Vetal

AI writes the code.Can you defend it?

Vetal questions you out loud about your own GitHub repo, then tests how you debug, build and review with an AI that makes planted mistakes. You get a scored report to show for it.

Start the free 5-minute round

Paste a public repo. No sign-in. See the four rounds

Four rounds. One report you can show.

Built for developers, not for employers. Pick a round to see what it looks like.

Defend your own code, out loud

Pick one of your GitHub repos. Vetal reads it, then questions you in a live voice call: why you made specific choices, and what would break if you changed them. It assumes any of the code may be AI-written. The question is whether you understand it.

Scored on: Understanding, Judgment.

Sample sessionRound 1, on your own repo
36def retry_with_backoff(call, tries=5, base=0.5):
37 for attempt in range(tries):
38 try:
39 return call()
40 except TimeoutError:
41 delay = base * attempt
42 time.sleep(delay)
43 raise RuntimeError("gave up after retries")

03:12VetalYou called this retry with backoff. Line 41 waits zero seconds after the first failure. What happens under load?

03:21YouOh. Attempt starts at zero, so the first retry is instant, and the delay only grows linearly. Under load every client hits the service again at once. I'd double the delay each time and add jitter.

Caught. Saved to your report as evidence, with the timestamp.

Finishing earns no points by itself. A partial solution you can explain beats a complete one you can't. After every round you see your weakest moment and a better approach, then you retry.

A report with evidence, not a badge.

  • A score out of 100, and how far you moved.
  • A level for each of five skills. Click one to see the quote that earned it.
  • Which planted mistakes you caught, round by round.
  • A share link. Transcripts stay private unless you share them.
Sample report, made-up dataPython backend track
78/100

“Attempt starts at zero, so the first retry is instant. Under load every client hits the service again at once.”

Evidence for Understanding, level 4. Round 1, 03:21
  • Round 1Explained
  • Round 2Caught
  • Round 3Caught
  • Round 4Missed

Code is cheap now. Thinking isn't.

Companies have started letting candidates use AI in coding interviews. The AI writes the code in seconds. What gets judged is everything around it.

Example task: delete a folderWritten by AI in seconds
@router.delete("/folders/{prefix}")
def delete_folder(prefix: str, user=Depends(current_user)):
files = Files.where(path__startswith=prefix)
files.delete()
25 audit.log("folder deleted", prefix)
26 return {"deleted": True}

Vetal scores these four moves, not the typing.

  1. FramingAsked first: should deleting a folder remove files other people shared into it?
  2. Steering the AIHad the AI write a test with two users before changing anything.
  3. Checking the AI's workCaught it: no owner filter, so "reports" matches every user's reports folder.
  4. JudgmentAsked for a soft delete, so a mistake can be undone.

Pay once. No subscription.

Prices in USD; tax is added at checkout where it applies. Paid rounds open Wed Nov 25.

Free try

$0

  • One 5-minute Round 1 on one of your repos
  • A score
  • Your single weakest moment

Job Hunt Pass

$99 for 90 days

  • Up to 6 core loops
  • Round 5, system design
  • Variants tuned to a job description you paste
  • Report refreshes

Questions

Is my code safe?

Secret-looking values, such as API keys, tokens and private keys, are masked before your code is analysed. Transcripts stay private unless you share them. The privacy notice lists what is collected and who processes it.

Can I use AI during the rounds?

Yes. In Rounds 2, 3 and 4 you work with an AI assistant or agent, and how you steer it and check it is what gets scored. Round 1 is a conversation about your own code.

How is it scored?

On what you understand and catch, not on whether you finish. You get a score out of 100 and a level for each of five skills, each with a quote and timestamp from your session as evidence.

What if I can't finish a round?

That is fine. Finishing earns no points by itself. A partial solution you can explain scores higher than a complete one you can't.

What languages can I use?

The free round and Round 1 work on a public repo in any common language. Rounds 2 to 4 start with a Python backend track; a full-stack TypeScript track follows in December.

Who sees my report?

Only you, until you share the link. You control what is public, and transcripts stay private unless you choose to share them.

Will employers accept it?

No employer endorses Vetal, and we won't claim one does. The report is evidence you can put in front of anyone: a score, quotes and timestamps they can read for themselves.

Why “Vetal”?

In the Indian Vikram–Betal tales, Vetal is the spirit who asks the king a riddle and punishes a wrong answer. The name also contains “vet”.

The next interview may let you use AI. Find out how you do first.

Start the free 5-minute round