What a resume screener really does
A resume screener triages a large applicant pool by scoring each CV against the requirements of a role, so a human recruiter spends attention on the most relevant candidates instead of reading every CV cold. The honest framing matters: it is a triage and surfacing tool, not a decision-maker. The simplest version, which the tool below implements, scores keyword coverage against a structured job spec — transparent and auditable. Production systems layer semantic matching on top, but the scoring logic and the fairness discipline are best understood from the explicit version first.
How it works
The pipeline has four stages. Extraction turns a CV file into plain text — in production a PDF text-extraction library with an OCR fallback for scanned or multi-column documents. Structuring parses the job description into weighted criteria: must-have skills (high weight) and nice-to-have skills (lower weight). Scoring checks how many required and preferred terms the resume covers and combines them into a weighted percentage. Reasoning produces a short, explainable rationale — which requirements were met and which were missing — so a human can act on the score with their eyes open.
The interactive tool below runs the scoring and reasoning stages live in your browser. Paste a set of required and preferred keywords and a resume, and it returns a weighted match score, the present and missing terms, and a shortlist recommendation. It is a demonstration of the mechanics, not a hiring authority.
Tips and the fairness layer
Keyword matching is the floor; the upgrade is semantic matching with embeddings or an LLM so “led a team of eight” satisfies “people management” without the exact words, and keyword-stuffing is penalised. But the most important layer is fairness, not accuracy. Score only job-relevant skills and experience — never names, photos, ages, schools, or location proxies for protected characteristics. Make every score explainable, audit outcomes across groups for disparate impact, and keep a human accountable for every decision. Automated employment decisions are regulated in many jurisdictions, so a screener that auto-rejects is a legal and ethical hazard. Build it to surface candidates, never to reject them on its own.