We use cookies for analytics and advertising to understand what's working and reach people who might find this useful. Essential cookies for sign-in always run. Privacy policy
What's behind FitToHire
A purpose-built extraction and matching engine — trained on real hiring data, not generic language models.
The scale
Real numbers from the production system.
26 domains
Every skill lives in a domain with its own vocabulary, relationships, and context rules.
191,538 total skills across all domains
The language layer
2.2 million phrases mapped to normalized skills. Watch how a resume phrase becomes a match.
825,000 trigger phrases in Tier A alone — covering every way people describe skills
The comparison
Every skill classified. Every gap identified.
How it works
Six stages from raw text to an explainable score.
Three-Mode Extraction
Three independent extractors — lexicon lookup, neural span detection, and full-text scanning — process every document. Results are merged and deduplicated.
Skill Normalization
191K skills across 26 domains with concept-key dedup. "AWS infrastructure management" and "managed AWS infra" resolve to the same canonical skill.
Cross-Encoder Disambiguation
When a phrase maps to multiple skills, a fine-tuned cross-encoder trained on 2M real job description pairs picks the right one from context.
Evidence-Based Scoring
Every matched skill links back to the exact phrase in your resume. No black boxes — every score point is explainable.
Requirement Weighting
Must-haves, preferred, and nice-to-haves are classified separately. A missing must-have weighs more than a missing nice-to-have.
Quality Gate
A diagnostics layer checks extraction quality before scoring. Low confidence gets flagged, not guessed.