Methodology
A deterministic lexical comparison—not an AI judgment.
Resume Keyword Screener compares language in supplied résumés with a job description. Skills-focused mode applies a documented curated taxonomy, phrase list, stop-word list, and synonym map. Full lexical mode preserves the original unique-token comparison. Relevant Keyword Review is an additive curated view that separates actionable technical and professional concepts from the raw Full lexical denominator without changing that score.
INTERACTIVE SYSTEM MAP
Follow the deterministic pipeline.
01Input
Files and text are bounded and validated.
02Parse
Supported documents become request-scoped text.
03Normalize
Documented tokens, phrases, and synonyms resolve.
04Compare
The protected Python engine computes lexical evidence.
05Review
Raw overlap and curated review signals render as separate factual views.
EXPLODED ANALYSIS CORE
One result, separated into accountable layers.
This diagram explains system ownership. It is architecture—not a résumé score, diagnostic, or live parser trace.
What the result means
Keyword Coverage describes overlap in lexical terms. It does not measure candidate quality, predict hiring outcomes, or reproduce a specific ATS.
Full lexical coverage is a raw compatibility baseline: every unique job-description token receives equal weight. Repeated occurrences affect the ordering of missing terms, not the coverage score. Generic words can therefore appear in raw matched or unmatched lists; they are labeled as raw terms rather than résumé skills or recommendations.
Skills-focused coverage gives every categorized job-description concept equal weight and reports Uncategorized concepts separately. The result states how many curated role concepts form the score denominator so a sparse taxonomy match cannot be mistaken for whole-document coverage.
Relevant Keyword Review is separate from both score denominators. It uses only explicitly curated categorized concepts and aliases, preserves exact source evidence, and marks each concept represented or missing. A missing concept is a review prompt only; it is not evidence that the résumé should claim experience the user does not have.
Normalization boundaries
Equivalences are explicit and regression-tested. The system does not infer related experience. For example, a Bradford assay does not imply protein purification, scale-up does not imply technology transfer, experimental design does not silently become design of experiments, and cell culture does not imply protein expression.
Limitations
- Curated normalization does not understand every synonym or context.
- Raw Full lexical overlap intentionally includes ordinary job-description vocabulary.
- Scanned PDFs require OCR, which is not included.
- Multiple résumés are combined into one comparison.
Language boundary
The current rules and curated taxonomy are English-first. Unicode text is handled safely, but the analyzer does not claim multilingual semantic understanding.