Hiring
AI recruiting software with psychometrics underneath
AI in recruiting has a credibility problem it earned. Tools trained on historical hiring data learned historical hiring bias, and the flagship failures of the last decade were all variations on that theme. The fix is not better models. It is not letting the model decide.
Interprets scores; it does not compute them
Models trained on your historical hiring decisions
On every score, flag and generated recommendation
Where the AI belongs, and where it does not
TalentSpark's scoring is psychometric, not generative. Trait estimates come from IRT models fitted to calibrated item banks. Percentiles come from normative samples. Role-fit comes from a trait-to-demand mapping derived from job analysis. None of that is a language model, and none of it is learned from your past hires.
Claude enters afterwards, to do the thing language models are actually good at: reading a scored profile in context and writing an explanation a hiring manager will act on. It interprets, it drafts interview probes, it flags contradictions between the profile and the role. It does not produce the numbers, and it cannot override a validity flag.
Why we do not train on your hiring history
A model that learns from who you hired will learn who you hired. If your engineering team is ninety percent one demographic, a model fitted to that outcome learns to prefer that demographic, and it will find proxies for it even when you strip the protected attributes — because everything correlates with everything.
This is not a hypothetical and it is not solvable with a fairness penalty term. It is a consequence of the training objective. The only reliable mitigation is to not set that objective, which means grounding the decision in instruments validated against performance rather than against your history.
What the AI actually produces
Each scored candidate generates a narrative report and a structured interview guide anchored to their profile. Where the profile shows an elevated Machiavellianism score, the guide contains probes about political disagreement with a rubric distinguishing influence from manipulation. Where Conscientiousness is high but its industriousness facet is low, the guide asks about follow-through on unglamorous work.
The job description optimiser rewrites postings for the traits the role actually demands and flags exclusionary language. The team optimiser models composition scenarios against trait complementarity. All of it is generated text, all of it is grounded in scores computed elsewhere, and all of it is labelled as advisory.
Regulatory posture
The EU AI Act classifies AI systems used in employment and worker management as high-risk, triggering obligations around risk management, data governance, human oversight, transparency and record-keeping. NYC Local Law 144 requires annual independent bias audits of automated employment decision tools, with published results and candidate notice.
Both regimes assume a human remains the decision-maker and that you can reconstruct why a decision was made. TalentSpark is built to make a human decision defensible rather than to make an automated one. Every score, flag and generated recommendation is logged with its inputs.
Frequently asked questions
Does the AI make hiring decisions?
No. It scores, explains and recommends. A human decides, and the record shows what they were shown. This is both a design position and, under the EU AI Act's human-oversight requirements for high-risk employment systems, an increasingly compulsory one.
Is TalentSpark an automated employment decision tool under NYC Local Law 144?
If you use it to substantially assist or replace discretionary decision-making, it falls in scope and you need an annual independent bias audit. TalentSpark produces the selection-rate data that audit requires. Whether your specific configuration is in scope is a question for your counsel.
Which model do you use?
Claude, for report generation, interview guide drafting and job description analysis. It reads scored profiles; it does not compute scores.
What happens to candidate data?
It is encrypted at rest with AES-256 and in transit with TLS 1.2+, is never used to train models, and is never sold. See the security and GDPR pages for data residency and retention detail.
Related
Candidate Screening Software
Screen candidates on validated psychometric evidence instead of résumé proxies. Role-fit scoring, adverse impact monitoring, and an audit trail behind every ranking.
Pre-Employment Assessment Tools
Pre-employment assessment built on validated psychometrics and structured interviewing. Score candidates on evidence, monitor adverse impact, defend every decision.
Adverse Impact Analysis Software
Monitor selection rates by protected group against the 4/5ths rule continuously. Bias audit evidence for NYC Local Law 144 and the EU AI Act, produced as you hire.
HR Analytics Software
HR analytics software that measures hiring quality, adverse impact, engagement and bench strength — grounded in psychometric data your HRIS does not collect.
Predict performance before day one.
Validated psychometrics and Claude AI, in one platform. No credit card required.