For hiring teams
Interview data and machine learning candidates with one link
Send a data or machine learning candidate one link. An AI interviewer asks about their real work, and you decide from the quoted scores.
Free while Aethel is in early access: one interview link, up to 25 interviews a month, about 3 of them with a video interviewer, the rest by voice. No card needed.
Hiring for data and machine learning
What makes this hard to screen
Many data resumes look alike: Python, SQL, a few model names. The hard part is telling who checks their data and tests honestly from who reports one accuracy number. You also want to know whether anyone ever used their work to decide something.
Roles this field covers: data scientist, data analyst, machine learning engineer, analytics engineer, data engineer, applied scientist, MLOps engineer.
Listen for these in a transcript
- Describes looking at the data before trusting it, and what they did about missing or skewed rows.
- Tests the model the way it will be used, such as a time-based split or a fair baseline, and says what the metric missed.
- Ties the work to a real decision and states the uncertainty in plain numbers.
Warning signs
- Drops awkward data without finding out why it was there.
- Reports results measured on the training data, or treats one accuracy figure as proof.
- Cannot say whether the analysis or model was ever used.
How it works
How to interview with a link
Step 1: Create a link for the role
Choose the field "Data and machine learning", the level, and a specialty such as data scientist or data analyst.
Step 2: Send it to your candidates
Share it wherever you already find data and machine learning candidates.
Step 3: Candidates interview on their own time
It takes about 15 minutes, and they know the interviewer is an AI.
Step 4: Review and decide
Mark each one Advance, Not a fit, or Undecided.
The interview
What the interviewer covers
It follows up on each answer and picks the data and machine learning topics that fit the role.
- A model or analysis that changed a real decision
- Messy, missing or biased data and how they handled it
- How they tested a model or analysis before trusting it and what the test missed
- A model or dashboard that broke or drifted after launch
- An experiment or A/B test and how they read the results
- Explaining uncertainty to people who wanted a yes or no answer
- Choosing between a simple approach and a complex one
- A pipeline, table or report other people relied on and how they kept it correct
What the scoring looks for
Strong work: the candidate checks the data before trusting it, validates the way the work will be used (held-out or time-based tests, a fair baseline), states uncertainty and limits, and shows the effect on a real decision with numbers.
Warning signs: the candidate drops inconvenient data without finding out why, tests a model on the data it was trained on, treats one accuracy number as proof, or cannot say whether anyone used the result. The scoring AI is told to score an area below 60 when its answers show one of these.
Your questions
Sample must-ask questions
Up to three per link. The interviewer is told to ask each one as written, and every result shows whether it was asked word for word, asked in other words, or not asked. Examples:
- Tell me about an analysis or model that someone acted on. How did you check it was right before they did?
- Describe a time the data was wrong or incomplete. How did you find out, and what did you do?
- How do you explain a result with real uncertainty to someone who wants a single number?
Starter templates
Start from a common role
Start a link from one of these and edit anything.
Analytics engineer
Focus: Data modeling in SQL, transformation tools such as dbt, data tests, shared metric definitions
Data analyst
Focus: SQL, dashboards and reports, checking data quality, explaining results to non-technical teams
Data engineer
Focus: Building data pipelines, batch and streaming data, data quality checks, warehouse design
Data scientist
Focus: Statistics and experiments, predictive modeling, SQL and Python, turning analysis into business decisions
Machine learning engineer
Focus: Training and evaluating models, putting models into production, watching model quality after launch
Results
What you see for each candidate
The full transcript
Every answer, in order.
“Not enough evidence”
Shown instead of a guess, never as a low score.
Your decision and private notes
Only your team sees them.
The four scored areas
Each out of 100, with quotes.
- Problem solving (messy data, failures, trade-offs)
- Counts as evidence when the candidate describes a specific data or modeling problem, its cause, what they tried, a trade-off, or a measured result.
- Modeling and analysis (methods, metrics, validation)
- Counts as evidence when the candidate describes how they built or evaluated a model, analysis or data pipeline: data sources, features, methods, metrics, validation, or how they checked the result was right.
- Communication (explaining findings and their limits)
- Counts as evidence when the candidate describes explaining findings or models to other people: a report, a dashboard, a presentation, or walking stakeholders through results and their limits. Answers that respond to the question asked, in a clear order, with a concrete example, also count, but only up to 80 on their own. Grammar, word choice and fluency are never judged.
- Collaboration (stakeholders, engineers, feedback)
- Counts as evidence when the candidate describes working with other people: stakeholders, engineers, product partners, reviews, mentoring or feedback. Saying "we" is not, by itself, collaboration evidence.
See a whole result the way your team would: the sample software and data result, with a made-up candidate.
Aethel never scores faces, voices or accents, and does not verify degrees, licenses or work history. How Aethel uses AI
FAQ
Questions hiring teams ask
Can it tell an analyst from an ML engineer?
Set the specialty on your link, for example data analyst or MLOps engineer. The questions then follow that kind of work: dashboards and experiments for one, pipelines and model drift for the other.
Does it run SQL or notebook exercises?
No. Candidates talk or type about past projects. A hands-on exercise, if you want one, belongs in a later round you run.
Can we ask about our own experimentation setup?
Write it as a must-ask question, for example how they would read an A/B test that ended with a flat result. The interviewer asks it in your words, and the result shows their answer.
See also the questions companies ask and the pricing page.
Send your first data or machine learning interview link
Want to show candidates what to expect? What the interview covers for data scientists, analysts and ML engineers
Interview candidates in other fields
- Software engineering
- Design
- Product management
- Finance and accounting
- Law and compliance
- Healthcare and life sciences
- Healthcare support and caregiving
- Science and research
- Writing, editing and languages
- Education and training
- Childcare and early education
- Sales and customer success
- Customer service and support
- Retail, food and hospitality
- Administration and office support
- Government and public service
- Nonprofit, church and community work
- Skilled trades and facilities
- Operations and other fields