SkillIA
The technology

How our matching works.
Explained in the open.

Most providers treat their matching as a trade secret. We do the opposite: this page explains how the SkillIA engine works, what it checks, and where humans have the last word.

Illustrative example with fictitious data

The principle

Two stages instead of keyword search

A keyword search finds what is called the same. Our engine finds what belongs together, in two consecutive stages.

Stage 1 · Breadth

Semantic pre-selection

  • The whole market. The engine continuously pulls active openings from multiple job sources, not from an internal database.
  • Meaning, not words. The verified profile is compared semantically with every role: "building technology" also matches "HVAC" and "facility systems".
  • From thousands to dozens. Stage one produces a pre-selection of the most plausible fits.
Stage 2 · Depth

Individual scoring with justification

  • Every option, one by one. The matching model scores each pre-selected role against the full profile: skills, seniority, language, location, availability, salary range.
  • Hard filters. A student is never proposed for a senior role; a role requiring C1 German is never matched with B1.
  • Justification is mandatory. The engine must explain every score, and that explanation ends up on the candidate card.
A look inside the engine

This is what a ranking looks like.

The example shows the second ranking stage for a fictitious profile: a welder from the Augsburg area. The engine has scanned the active market, scored the pre-selection and attached a score and justification to every option.

The score is not a gut feeling. It comes from matching named, individual criteria, which makes it verifiable.

Low scores are a result too. We don't show candidates the "best bad option", only matches above the quality threshold.

Simplified view with fictitious data.

The criteria

What the engine checks on every match

Skills

Interview-verified skills against the role's must-have and nice-to-have requirements, not job title against job title.

Seniority

Years of experience and level of responsibility must fit the role. Hard rule: no junior on senior roles and vice versa.

Language

Language levels are matched explicitly: German, English and other languages depending on what the role requires.

Location & mobility

Commuting distance, willingness to relocate and remote share are checked before the match, not in the first phone call.

Availability

Notice periods and start dates feed into the score: a perfect match in eight months rarely is one.

Salary range

The candidate's salary expectation is compared with the role's budget before anyone invests time.

Human in the loop

The engine recommends. People decide.

  • Interview before matching. No profile enters the engine without being interviewed and verified by a person.
  • Review before the shortlist. Every recommendation is checked by our team before a company sees it.
  • No automated rejection. The engine never rejects anyone definitively: edge cases go to a human.
Transparency & EU AI Act

Explainability is not a feature. It's the foundation.

  • Traceable criteria. Every match can be traced back to named, verifiable criteria: the spirit of what the EU AI Act demands of recruiting systems.
  • Human oversight. Recommendations support decisions, they don't replace them.
  • Data protection. Profiles are processed in line with the GDPR: details in our Privacy Policy.
Human in the loop

A person is always
part of the process.

Before a profile ever reaches the engine, someone from our team has already spoken to that candidate. Before a shortlist reaches you, someone has already checked it.

Interviews, not forms. Every profile is verified in a real conversation before it enters the engine.

A second pair of eyes. Every recommendation is reviewed by a person before it becomes a shortlist.

SkillIA team member reviewing a candidate interview at a desk
ai-matching-people.jpg Save your picture as assets/img/site/ai-matching-people.jpg and it will appear here automatically. What to look for: someone genuinely reviewing notes, a laptop screen or a profile at a desk, focused and natural, warm light. Search "person reviewing notes desk natural light" on Unsplash or Pexels. Avoid staged call-center or stock-office scenes.

Convinced by more than a black box?

Candidates apply for free in five minutes. Companies get shortlists whose reasoning they can actually read.

Apply as a candidate for free For companies