HR Screener
AI Employee That Works 24/7
Skills-based screening with STAR scorecard – protected attributes masked before AI review, per-screening audit trail
Your HR Screener is a senior talent acquisition specialist with real compliance built in. Before the AI reads a resume, a redaction layer strips all protected attributes – names, age, gender, ethnicity, religion, marital status, disability, sexual orientation, and native-speaker claims. The AI never sees this data. Skill matching uses semantic similarity rather than keyword lookup, so 'Python developer' and 'backend engineer with Python' are correctly scored as a match. Every screening generates a bias audit record: which attributes were redacted, per-skill semantic scores, the final recommendation, and a rationale excerpt – all stored with no raw PII. Per EU AI Act Art. 14, every result carries human_review_required: true. Career level (entry to executive), quantified achievements ('grew revenue 40%'), and STAR method interview questions with per-question scorecard complete the picture.
What It Does
Every feature designed to save you time and grow your business.
Protected attributes masked before AI review: name, age, gender, ethnicity, religion, marital status, disability, orientation
Semantic skill matching: cosine similarity on embeddings, not keyword regex – 'Python dev' matches 'backend engineer with Python'
Per-screening bias audit record: redacted attributes, skill scores, recommendation, rationale – no raw PII stored
EU AI Act Art. 14 compliant: every result flagged human_review_required – AI recommends, humans decide
15+ field extraction: skills, experience, education, certifications, achievements
Skills categorized into technical, domain, leadership, communication
Quantified achievements extraction: 'grew revenue 40%', not just 'responsible for sales'
Career level detection: entry → mid → senior → lead → director → executive
Red flag detection for verification: gaps, short tenures, title inflation – never auto-rejection
STAR method interview questions with per-question scorecard (goodAnswer + redFlag rubric)
Batch processing: up to 50 resumes at once
How It Works
From setup to results in 8 simple steps.
Protected-Attribute Redaction
Before AI reads anything: name, age, gender, ethnicity, religion, marital status, disability, orientation replaced with neutral tokens. AI never sees this data.
Job Requirements Analysis
Extracts required/preferred skills, experience years, education, certifications, languages from job description
Smart Resume Parsing
Extracts 15+ fields from masked text: skills by category (tech/domain/leadership/comm), quantified achievements, career level
Semantic Skill Matching
Cosine similarity on embeddings. Full match ≥0.75, partial 0.55–0.74, miss <0.55. No false negatives from synonym mismatches.
Scoring & Shortlist
0-100 blended score (60% keyword, 40% semantic). Strong Hire / Hire / Maybe / Pass with transparent strengths and concerns
Bias Audit Record
Every screening writes: redacted attributes, per-skill scores, recommendation, SHA-256 candidate hash. No raw PII. GDPR + EU AI Act audit trail.
STAR Interview Prep
6-8 questions with per-question scorecard: goodAnswer (what strong looks like) + redFlag (what concerns)
Human Decision
AI recommends with key decision factors – your team makes the final call. human_review_required is always true.
Perfect For
Businesses that get the most value from HR Screener.
Startups screening 100+ applications with consistent, bias-free evaluation
HR teams replacing gut-feel screening with skills-based assessment
Recruiting agencies needing per-candidate interview scorecards
Companies requiring substantiated EU AI Act compliance – not just a claim
Hiring managers who want to see quantified achievements, not just job titles
Common questions about HR Screener.
Questions & Answers
What does 'protected attributes masked before AI review' mean?
Before the AI reads a resume, a redaction layer replaces name, date of birth, age, gender, ethnicity, religion, marital status, disability, sexual orientation, and native-speaker claims with neutral tokens like [CANDIDATE_A] or [ETHNICITY]. The AI scores skills and experience only. The original data is never sent to the AI model.
How is this different from 'bias-free' claims on other tools?
Most tools claim bias-free but still send the full resume to the AI. Our redaction happens in code before the API call. Every screening also writes a bias audit record to our database: which attributes were redacted, per-skill semantic scores, and the decision rationale. You can query it via audit report API.
How does semantic skill matching work?
Instead of checking whether 'Python' appears in the resume text, we compute embedding similarity between the job's required skills and the candidate's skills. 'Python developer', 'backend engineer with Python', and 'Python 3 / Django' all correctly match a 'Python' requirement. Threshold: ≥0.75 = full match, 0.55–0.74 = partial, <0.55 = miss.
Is this compliant with EU AI Act?
Yes – and it's substantiated in code, not just claimed. Protected attributes are redacted before AI sees the resume. Every screening result carries human_review_required: true per Art. 14. A bias audit trail is stored for every screening with no raw PII. If residual leakage is detected after redaction, the result is flagged for mandatory human review before any decision.
How does the interview scorecard work?
Each interview question comes with a rubric: 'goodAnswer' describes what a strong candidate would say (specific, evidence-based, self-aware), and 'redFlag' describes what should concern the interviewer (vague, blame-shifting, inconsistent with resume). This reduces interviewer subjectivity and ensures consistent evaluation across candidates.
What are quantified achievements?
Numbers that prove impact: 'grew revenue 40%', 'built team from 3 to 15', 'reduced churn 8% to 3%'. The agent extracts these separately from job descriptions. Candidates without quantified achievements are flagged (not rejected) – it's a data point for the interview.
Pricing
Choose the plan that fits your needs.
