The interview is already in your job spec. We just add the cloud.

Assess engineers by building with them.

You and your candidate get a real problem in a real AWS, Azure or GCP environment — a dedicated account, provisioned in minutes, spend-capped, torn down when you're done. You work the problem together. Afterwards the candidate can get the same report as you — every judgement in it links back to something they actually did.

For teams hiring DevOps, SRE, platform, MLOps, database and full-stack engineers

real account, not a simulator · spend-capped · torn down & verified empty

Session notes: PriyaPlatform engineer · GCP · 41 min

The environment is up: a real GCP project and a Terraform repo. The CI service account has roles/editor on the whole project. The task is to get it down to least privilege without breaking the deploy.

Before she touches any Terraform, she pulls the audit log to see what the pipeline actually calls.Went to the evidence first.

Asks whether Cloud Build should run as this account or impersonate it. We talk it through, and she picks impersonation and explains why.

Removes editor. The deploy check fails, so she reads the 403 and adds run.invoker on the one service that needs it.Didn't guess. Read the error.

Narrows storage.objectAdmin from the whole project to the single artifacts bucket.

Opens a merge request with a description I could hand to a reviewer as it stands. All five checks pass.

Session closed and the project torn down. Cloud spend: $1.84.

An illustrative session. The candidate gets these notes too.

See the session we'd run.

Paste a CV and we'll show you the assessment plan a real session would run: the skill profile we infer, the hands-on tasks we'd set, and the questions we'd open with. Free, ~20 seconds, nothing leaves your browser except the CV text itself.

The candidate's CV

Paste the text (a PDF's text works too — copy it out).

checking availability…

How a session runs

A Q&A shows you how someone talks about the work. Here you watch them do it. The reason companies don't already: not the format — the environment. So pairdojo provisions it for you, and takes it away afterwards.

A dedicated cloud account or project, isolated from your infrastructure, spend-capped, torn down and verified empty afterwards. Your IT team's involvement is zero. Every hint and task is logged, so a helped answer reads as a helped answer.

From these fundamentals we build the session's real environment.

A six-mat floor plan. The candidate to the left with their CV and their abilities; the interviewer to the right with the role spec and questions; the problems we set and what the candidate achieves fill the middle. Their CVwhat they say they can do Your role specwhat the job needs Candidatehow they work Problems we settailored Intervieweryour bar What they achievecommits, checks, the report
  1. Send us the role and the CV

    We match them and suggest problems that probe the gaps. Swap any of them for your own. The same session format runs just as easily for MLOps, database and full-stack roles. Optionally bring your own model inference and submit problem ideas via our API.

  2. We build the environment

    That's a dedicated cloud account or project with the editor, git, the CLI and the broken infrastructure already in place. It takes 3–15 minutes, and you set the spend cap and time limit.

  3. You work it together

    Share your screen on whatever call you already use. Pair on it, or sit back and ask questions. Either way, everything gets recorded.

  4. Both of you get the report

    Once the environment is torn down, you each get the same report, built from what happened in the session.

A report you could defend in a hiring meeting.

Every level in the report points to something the candidate did: a commit, a command, a moment in the recording. If the session didn't touch a skill, the report says so rather than guessing.

The problems are the kind of thing the job involves — an over-broad IAM policy, a failing Helm release, a migration that locks a table — not puzzles. And they get the same report you do: same evidence, same wording, something a strong candidate can keep.

If they used AI tools, the report says that too, and shows how. You decide what that means for the role.

Levels describe depth, not years:
1 · Can do
gets it working
2 · Idiomatic
does it the way the platform intends
3 · Robust
anticipates failure and blast radius
4 · Architect
shapes how other people will do it

Priya N.

Platform engineer · GCP · live pair, 41 min

74across 3 evidenced skills
Security & IAM Robust Replaced a project-wide editor grant with three scoped roles without breaking the deploy.MR !1 · commits 3f1c2e0, a91d4b7
Terraform Idiomatic Clean resource-level bindings, and the plan was read before every apply. Hard-coded the project ID twice.commit a91d4b7 · plan at 00:18, 00:26
CI/CD Robust Moved Cloud Build to impersonation rather than a long-lived grant. When the deploy failed, she found the missing role from the 403 in four minutes.commit 3f1c2e0 · recording 00:12, 00:19–00:23
Networking Not seen This session didn't touch networking.

Interviewer's noteAsked the one question that changed the design. I'd happily have her on call with me.

Cloud spend $1.84, itemised. The candidate gets the same report.

Two ways to run it

Live pair

Together, in real time

You're both in the environment at the same time. It's closest to a normal working day, and the best way to find out what someone is like to work with.

Brief, then walkthrough

24–72 hours on their own, then an hour with you

They get the environment, the brief and a budget, and work when it suits them. Then they walk you through what they did and why, and you ask about the decisions.

Also available solo — against a spec, for engineers testing themselves. The same evidence-backed report, on your own clock.

Pricing

$50 per candidate, per session. Plus their actual cloud usage, doubled — the markup covers guaranteed teardown and any stray bill, which is ours, not yours. You set the spend cap per session; there are no seats and no annual commitment, and the candidate never pays anything.

A typical live session: $50 plus a few dollars of cloud.

Per candidate$50
Cloud billactual usage, ×2 — capped by you
Spend capyou set it per session
A typical live session$50 plus a few dollars of cloud
CloudsAWS, Azure, Google Cloud

Try it on someone you've already hired.

Choose three to five engineers on your team who are happy to take part. We'll run sessions with them and send you the reports. If the reports don't match what you already know about those people, you've lost an afternoon.