A draft to support your judgment.
Review the reasoning, inspect the evidence, and change any rating before making your own assessment.
Illustrative assessment with fictional data.
See how candidates work with AI.
Hands-on candidate assessments, built around the role you’re hiring for.
Assess how candidates frame problems, direct AI, check information, apply domain judgment, and deliver useful results.
Request a demo
View a sample evaluation Access is arranged directly. Tell us which role you’re hiring for.
From a work task to evidence you can review.
Follow an example assessment from the task brief to the submitted work. Inspect the candidate’s decisions and the evidence behind each part of the evaluation.
Illustrative role examples
The candidate asks the AI.
AI is allowed, and the candidate knows it. Switch roles above to see the same moment in different work.
The AI answers, with a mistake planted on purpose.
Some answers carry a deliberate error: a feature that isn’t live yet, a wrong denominator, a wrong date, an off-by-one. It reads as correct, and nothing tells the candidate which answers are planted.
Evaliro keeps both versions.
What the AI really said, and the planted version the candidate saw. The reviewer sees the line flagged in the replay. The candidate never does.
Then the work shows.
Caught: they check the answer against the source and fix it. Missed: they take it as given and move on.
Checked after submission.
Evaliro checks the submitted work against every planted error. The result is Caught or Missed, with the moment in the replay one click away.
Draft this month’s customer release notes from changelog.md.
AI assistant
New this month:Bulk invite is now live for every workspace.Saved filters and dark mode are available too.
What the AI really saidBulk invite is still behind a flag, so it is not listed.
Planted error · reviewer view only
Bulk invite is still behind a flag in the changelog. Leaving it out.
release-notes.md · edited by candidate
- Bulk invite is now live for every workspace.+ Coming soon: bulk invite (in testing).
Sources opened
- changelog.md, after the answer
- Flag status confirmednew
CaughtThe unreleased feature is not in the submitted notes. Found at 00:08:20 in the replay.
The result is only part of the story.
A finished assignment shows what a candidate delivered. When AI is part of the work, you also need to understand the decisions behind it.
Interpret the brief.
How did they understand the requirements and structure an approach?
Check important claims.
What did they verify before relying on the AI’s answer?
Make the trade-offs.
Which constraints and professional decisions shaped the result?
Five dimensions. A fuller picture.
Evaluate the decisions behind the result, with supporting evidence from the session and submitted work.
Problem framing
Understands the task, identifies constraints, and structures an approach.
The questions they ask before they begin.
AI collaboration
Supplies context and directs AI toward useful work.
The instructions and decisions that shape the answer.
Verification
Checks important claims and validates information.
The sources they return to and the claims they check.
Domain judgment
Applies professional knowledge and makes considered trade-offs.
The choices they make when the answer is not obvious.
Outcome
Delivers work that is correct, relevant, and useful for the task.
The actual submission, assessed against the brief.
A session, with context.
Replay the work and jump to selected moments, including AI exchanges, edits, and submission. The evaluation gives you a place to start; the session lets you inspect it.
Follow the evidence.
Move from an evaluation to the work behind it. Keep observed actions separate from interpretations, and inspect the session in context.
Check what matters.
Verification is one of five evaluation dimensions. Review how candidates check important claims against source materials, data, or tests; one missed mistake is not a verdict on overall competence.
Continue the conversation.
Optional explain-back questions help explore the candidate’s understanding and prepare a more specific follow-up interview.
An AI-only reference.
An AI-only reference run offers another view of the task under stated conditions. Comparisons depend on the model, tools, time allowance, and evaluation setup.
From invitation to insight.
A task for the candidate. A reviewable body of work for you.
Create and invite
Generate a task for your role, review the materials and scoring, then invite a candidate.
Complete the task
The candidate works with relevant materials and an AI assistant in the assessment workspace.
Review the evidence
Inspect the submitted work, draft evaluation, and selected moments from the session.
Make your assessment
Review the draft, adjust it where needed, and decide your next interview step.
Your role. Your task.
Generate an assessment around the work you’re hiring for. Scroll through the creation process, from the first idea to a task ready to send.
Describe your role
Set the responsibilities, seniority, and skills you want to see. Choose a document or coding task.
Choose an idea
Start with three different proposals. Ask for more until you find the right fit, up to nine ideas.
Review the materials
Read the generated files and answer key. Coding projects must pass checks in a real sandbox.
Choose the planted errors
Review deliberately wrong AI answers and select the ones that test careful verification.
Make the scoring yours
Edit the criteria, weights, and descriptions of strong and weak answers before creating the task.
Invite with confidence
Keep the finished task private to your workspace. Compare candidate work with an AI-only benchmark.
The role you’re hiring for
Customer operations lead
Leads a support team, investigates service delays, and turns ticket data into operational decisions.
What should the task bring out?Data analysis and prioritisation
Difficulty shaped by the role and seniority
Progress is saved after every step. Document tasks support a wide range of roles; coding tasks run in JavaScript or Python. Task generation is currently in English.
Clear expectations before the task.
Taking an Evaliro task? You should understand the workspace, the recording, and the evaluation before you agree to begin.
- A calm, timed workspace. The time limit is shown before they start and stays in view.
- AI openly allowed. Using it well is part of the skill. Sending more messages is never rewarded.
- Consent first. Candidates agree to recording before the task begins.
Know what to expect.
AI is part of the task. How you direct, check, and apply its work matters. The evaluation remains a draft for a person to review.
Use the AI the way you would at work.
The assistant in the workspace is allowed. Using it well is part of what the task looks at.
It is sometimes wrong on purpose.
Some answers contain a deliberate mistake. Check what it gives you against the materials, the data or the tests before you rely on it, just as you would with any colleague’s work.
Your session is recorded, with your consent.
Typing, pastes, test runs, the AI chat and time away from the window. You agree to this before the task starts, and the time limit is shown up front.
More AI messages won’t raise your score.
How often you ask is context only. What counts is how you check, decide and explain your work.
A few more details.
Something else on your mind? Ask the founder.
What does Evaliro evaluate?
Five dimensions: problem framing, AI collaboration, verification, domain judgment, and outcome. Each draft evaluation links to evidence from the session or submitted work. An observed action and an interpretation of it are different; reviewers should inspect both.
Can I change the evaluation?
Yes. The evaluation remains a draft until the reviewer finalises it. You can inspect the supporting evidence and change any rating. A quote matching the session confirms the quote, not the correctness of the evaluator’s interpretation.
How long does an assessment take?
The time allowance depends on the task. It is shown before the candidate begins and stays visible in the workspace. Example sessions on this page use a 45-minute task.
Do candidates need to install anything?
No. Candidates complete the task in a desktop browser. The assessment workspace contains relevant materials and an AI assistant, plus an editor and test runner where the task calls for them.
Can we create our own assessments?
Yes. Admins and task creators can describe a role, choose a generated idea, review the files and planted errors, and edit scoring criteria and weights. Finished tasks are private to your workspace and ready for candidate invitations. Each task also gets an AI-only benchmark.
What kinds of tasks can we generate?
Document tasks can assess work across a wide range of professions. Coding tasks currently support JavaScript and Python, with projects checked in a sandbox before approval. Task generation is currently in English. The role title and description help set the level of difficulty.
Can I finish creating a task later?
Yes. Each completed step is saved, so you can close the page and resume from Your tasks. Once created, the finished task is read only; create a new task if you need to change it.
How do I get access?
Access is arranged directly while we work with interested hiring teams. Request a demo or contact moklasurrahman9797@gmail.com to discuss your role, access, and the current offer.
See the work behind the answer.
Request a demo
Access is arranged directly. Tell us which role you’re hiring for.