Hire AI engineers who can get it running.
There is the AI you can talk about and the AI you can get to run on a machine. Candidates fine-tune a model and stand up a retrieval pipeline in an isolated cloud environment, and what comes back is whether it works, not how well they described it.
$ docker compose up -d
3 services started
$ systemctl status api
active (running)
$ curl -s localhost:8080/health
{"status":"ok"}
Live environment
Running- vmlab-vm-01Ready
- svcapi-serverReady
- lbgatewayProvisioning
2h 12m left of the 3h limit
Talking about models and shipping one are different jobs
Without Scalyz
With Scalyz
What is at stake
The job is not the demo, it is the day the demo breaks
The hard part of an AI role is rarely the idea. It is making it run, and keeping it running, on infrastructure that fights back.
Where it breaks
Between notebook and server
Plenty of candidates can make something work in a notebook. The role is what happens when it has to run somewhere else.
What an interview misses
The operational half
Model intuition is easy to talk about and hard to verify. Getting the thing to actually run is easy to verify and hard to fake.
What you are hiring for
It runs on Monday
You are not hiring a description of a pipeline. You are hiring the person who gets it up and keeps it up.
Live in an afternoon
Three steps from the job post to a shortlist you can defend, with no screening calls.
Backend hiring, July
Open
6 invited · 4 completed
Closes Friday, 18:00
Invite the whole list at once
You pick the test and the deadline, then send your whole applicant list at once, by email, CSV, or a shareable link.
Live environment
Running- vmlab-vm-01Ready
- svcapi-serverReady
- lbgatewayProvisioning
2h 12m left of the 3h limit
Candidates work when it suits them
Each candidate gets their own machine, a real one, with a real task already loaded.
Camille Durand
DevOps assessment
No integrity flags
The shortlist ranks itself
Scoring runs automatically against the task you set, so the ranking builds as candidates finish.
$ docker compose up -d
3 services started
$ systemctl status api
active (running)
$ curl -s localhost:8080/health
{"status":"ok"}
Mission
3h limitShip it behind a load balancer
- Set up the web server
- Deploy the API service
- Route traffic through the load balancer
- Prove the health checks pass
2 of 4 complete
We test the half of the job that actually runs
The published missions here are operational: fine-tuning a model, and building a retrieval pipeline with LangChain. Both run in a real environment on the candidate's own machine, and both are scored on what the system does at the end. We do not grade research or model theory, we grade whether it works.
- Every mission runs in an isolated cloud environment, one per candidate
- Scored automatically on whether the system runs at the end, not on an answer
- You can import your own scenario as a .zip and run it yourself before you send it
The published missions for this role
The questions we get
Straight answers on how the tests, the scoring and the credits actually work.
What is a Scalyz lab, exactly?
A real virtual machine with a real mission, not a quiz. The candidate works with the same tools they'd use on the job, and the session is recorded so you can review the evidence later.
Who scores it?
Automatically. The checks run against what the candidate delivered, and you get a report broken down skill by skill, so no one on your team grades a thing by hand.
Can they cheat with AI?
A chatbot can suggest commands, but it can't run a live system for the candidate. The mission has to be done on the machine itself, and any cheating signals come with context, so you can judge them yourself.
What do I pay?
In credits: one per candidate. A credit is reserved when you send an invitation and released if it expires unused. You pay per candidate tested, never per user.
Can I bring my own exercise?
Yes. Upload your files as a .zip, rename the mission to match your role, and run it yourself for one credit first. What you tested is exactly what candidates get.
What is it like for candidates?
On their own time. They start when it suits them, work with real tools on a real machine, and finish in one sitting. It runs in English and French.
Test the AI work that has to run
Book a demo and we will run one of these missions on your own open role.
One credit per candidate, reserved on send, released on expiry
