For Chief People Officers & Heads of Talent Acquisition
A defensible hiring record - without adding a regulated AI decision tool to your stack.
Interview Kit Generator turns a job description into a structured, EU-bias-screened interview kit - competencies, behavioural questions, probes, and an anchored rubric - screened against equality-law risk before your panel ever sees it. The decision, and every score, stays with your people.
Built for multi-jurisdiction, high-volume employers in litigation-prone sectors - finance, healthcare, retail, large tech - where this is funded from the legal and risk budget, not the HR-tools budget.
The exposure doesn't move with the news cycle
Indirect discrimination is a live exposure, today
Liability under the EU equality directives applies in every member state right now, independent of how the AI Act's timeline moves. The exposure doesn't track regulatory appetite.
Your screening tools are now the liability surface
The EU AI Act classifies recruitment and candidate-evaluation AI as high-risk - attaching documentation, logging, and human-oversight duties to tools you may already be running.
When a claim lands, you need a record you don't have
The question is always the same: was this hiring job-related, consistently applied, and objectively justified? For interviews specifically, most employers have nothing to show.
What it does
Live today
- Paste a job description, get a structured kit back in under a minute - no account, no setup.
- Competencies named against an ESCO-style taxonomy, not invented fresh each run.
- 1-3 behavioural/situational questions, follow-up probes, and an anchored 1/3/5 rubric per competency.
- Every question screened against a versioned EU bias-screening ruleset, with flags and safer rewrites.
- Blank, structured fields for your interviewers to record their own scores.
- Copy-to-clipboard and JSON export to take into the room or your ATS.
Building next
- A dated, versioned, attributable record per candidate and per role.
- Adverse-impact monitoring across interview scores, so 'defensible' holds in production.
- Counsel-ready export for claims and data-protection inquiries.
- An org-wide competency framework and rubric, reused across every kit and every manager.
What it produces, and what that supports
Competency-to-role tracing, anchored to ESCO
Supports the objective-justification (job-relatedness) argument under EU equality law.
One rubric, applied to every candidate for the role
Your consistency defence against a disparate-treatment claim.
Human, multi-rater scoring
Human-in-the-loop - not a solely-automated decision under GDPR Article 22.
Jurisdiction-aware question screening
Avoids facially unlawful or indirectly discriminatory questions before they're ever asked.
Dated kit with a referenced screening-ruleset version
The first layer of the paper trail counsel asks for - the full per-candidate record is next.
We generate questions. Humans make the call. That's the whole boundary.
The system generates questions, probes, and rubrics for a human interviewer to use. It does not score, rank, or shortlist candidates - and it never will. That's not a feature gap we'll close later; it's an architectural commitment. It keeps the product off the EU AI Act's high-risk classification for automated candidate evaluation and outside GDPR Article 22's regime for solely-automated decisions with significant effects. More practically, it means the decision - and the responsibility for defending it - stays exactly where it already sits: with your hiring team.
“The moment a tool scores or ranks candidates, it becomes the regulated thing it was meant to help you avoid.”
Questions your team will ask before you do
Isn't this just a ChatGPT wrapper?
Generation is the on-ramp, not the product. The value is enforced consistency across every interview, jurisdiction-aware screening of every question, and - next - a litigation-ready record. None of that comes out of a chat session, and none of it persists in one.
Doesn't our ATS already do this?
Your ATS stores outcomes. It doesn't enforce a single scored rubric, screen questions for unlawful content, or produce a record of a structured human process. We sit on top and fill in the interview layer most ATSs leave blank.
We already run structured interviews.
Do you run the same rubric every time, screened for unlawful content, scored by multiple raters, with a record you'd hand to counsel on request? Consistency and the artifact are usually the gap - and they're exactly what a claim probes first.
Why does this matter if the AI Act's timeline keeps slipping?
The AI Act's deadlines are a moving target. Indirect-discrimination exposure under the EU equality directives is not - it's in force today, in every member state. The exposure is the constant; the cost of assembling a defence after the fact is the variable we remove.
Where we draw the line
This category is full of overclaiming. Here's what we won't say, even when it would make for a better slide.
No safe harbor
No EU equality directive offers one, for any selection method - structured, screened, or otherwise.
Not a validation service
This supports an objective-justification argument. Formal validation of the competency and rubric methodology is a job for your counsel and an I/O psychologist.
Not legal advice
Responsibility for the hiring decision - and its defence - stays with your organisation, exactly as it does under the AI Act (as deployer) and GDPR (as controller).
No “AI-Act-ready” badge
Standards and enforcement timelines are still unsettled. We reduce risk and help produce evidence; we don't certify outcomes.
What has to be true before this is load-bearing
- An I/O-psychologist-validated competency and rubric methodology.
- EU employment-counsel review of the screening ruleset and record format - current ruleset is a draft (v0.1), unreviewed.
- Adverse-impact monitoring running on real interview scores, not just demoed.
- Named experts behind the methodology, published alongside the product.
See it on one of your own roles
Paste a real job description and get a structured, screened kit back in under a minute.
Try the generator