We use analytics cookies to understand how this site is used. They
are optional and nothing here stops working if you decline — we set
no advertising or cross-site tracking cookies either way.
Cookie Policy
Expertini ATS combines full-scale semantic understanding with deterministic Candidate Match Scoring™ to evaluate candidates based on evidence—not keywords. Every recommendation is explainable, reproducible, and remains under human control.
Built for HR/recruiters who value evidence, consistency and transparency.
Since 2008 · independently operatedDeterministic scoring, not a black-box guessEvery score is auditable and evidence-cited
The Expertini Proprietary CMS Score
96%
Michael J.
Senior Full Stack Developer
Matches meaning, not just matching words.
Full-Scale Semantic UnderstandingReads CVs for meaning & evidence — not keyword bingo
AI-Assisted UnderstandingUsed only for language understanding — never to assign a score
Deterministic ScoringSame inputs.Same evidence. Same score. Every time.
Humans DecideThe score informs your judgement — it never replaces it
▶ Watch the 30-second overview — no sign-up needed
0Live jobs indexed ▲ growing daily
0Monthly job seekers ▲ network-wide
0Country job platforms ▲ 40+ specialist boards
0Deterministic scoring ▲ same inputs, same score
HumanEvery final decision ▲ the score only informs
HOW EXPERTINI ATS WORKS
Transparent. Reproducible. Defensible.
Language understanding where it helps, plain arithmetic where it matters, and a human making the call at the end. Every score can be traced back to the evidence that produced it.
STEP 1 · UNDERSTANDING THE ROLE
Read the Role
Before anyone is scored, we make sure the role itself is understood — in the employer's own words, not a template's.
Every requirement is identified and weighted by how the employer phrased it — "must have" and "required" carry the most weight (JRIS 90–100), "preferred" sits in the middle (50–70), "a plus" the least (30–50)
Hard blockers are flagged separately, so a missing mandatory qualification is never averaged away by strong scores elsewhere
The weights are visible and adjustable — nothing is inferred behind your back
Write naturally; there's no special syntax to learn
Tip: the clearer your must/preferred/a-plus language, the sharper your shortlist. Our explainable recommendations drafting assistant writes in exactly this structure.
STEP 2 · UNDERSTANDING THE PERSON
PII Anonymisation
A CV is a person's story. We extract the evidence — and remove everything that could bias a reader before analysis begins.
Skills, experience, education and certifications are extracted from PDF or DOCX CVs
Names, age, gender markers, photos and location identifiers are stripped before any semantic analysis — the matching engine never sees them
The full original CV stays available to your human reviewers, exactly as submitted
GDPR by default: candidate data is deleted after 24 months, and candidates get a real account on their local Expertini site to manage it
Ranking is based on evidence of skills — never on name, age, gender, or where someone lives.
STEP 3 · FULL-SCALE SEMANTIC UNDERSTANDING
Match on Meaning
Traditional keyword search rewards whoever games the buzzwords. Semantic understanding finds people who have actually done the work.
"Led a Django migration serving 2M users" matches a Python requirement — even if the word "Python" never appears
Each requirement gets a Candidate Skill Score (CSS) backed by the specific CV passages that evidence it
Language understanding is the only place Artificial Intelligence is used — it reads and relates text; it never assigns the score
Every match links back to its evidence, so you can verify any claim in seconds
To our knowledge, no other ATS runs semantic matching at full scale with the score kept fully deterministic.
STEP 4 · DETERMINISTIC SCORING
Score with Mathematics
The number that ranks a human being should be mathematical you can check — not a model's opinion.
CMS = Σ (CSSi × JRISi) / Σ JRISi
One published formula: each skill score multiplied by its requirement weight, summed, and normalised — that's the whole calculation
Same CV + same job description = the same score, every single time, forever
The methodology is submitted to an open research repository for researchers to evaluate, and openly documented — auditors, legal teams and candidates can all read exactly how a score came to be
No hidden model weights, no drift, no "the algorithm decided"
The full methodology is published at research.expertini.com — "From Stochastic to Deterministic" (Expertini Research Team, 2026).
STEP 5 · THE HUMAN CALL
You Decide
The CMS ranks and explains. Your team interviews, judges, and hires. That order never reverses.
Candidates arrive ranked with dimension-level breakdowns — see exactly which requirements each person evidences, and how strongly
Disagree with a score? Every input is visible, so you can see why and act on your own judgement
Your team's stage moves, ratings and interview feedback live alongside the score — the system informs, people decide
A complete audit trail records every decision, ready for any compliance review
An honest tool makes your judgement faster and better documented. It doesn't replace it.
LIVE · CMS ENGINE SANDBOX
Drag the weights. Watch the mathematics.
This is the exact formula running in production: CMS = Σ(CSS × JRIS) / Σ(JRIS). JRIS is how critical the employer says a requirement is; CSS is the evidence found in the anonymised CV. No AI touches the number — try to make it hallucinate. You can't.
Python & Distributed SystemsMUST HAVE
Cloud Architecture (AWS/GCP)REQUIRED
Team LeadershipPREFERRED
Open-Source ContributionsA PLUS
CMS SCORE
0%
—
Hard blocker active — a mandatory requirement has zero evidence. CSS pinned to 0; strengths elsewhere cannot mask it.
Σ(CSS × JRIS) = 0 Σ(JRIS) = 0 CMS = 0 / 0 = 0
Identical inputs → identical score, every single run. This transparency ships on every candidate's audit PDF.
Drag a candidate into Shortlist or Reject — or tap ✓ / ✗. The Deterministic Engine explains its reasoning live, using the same rule matrix and evidence behind that score.
Deterministic Engine showing: Sophia M.
Hard requirements for this job posting — pass/fail gates, independent of overall score
100%
Rule Compliance
Rank → Decide
SMART RANKING · drag a chip, or use ✓ ✗ ⇪ — ⇪ loads that candidate's evidence into the sandbox above
Expertini CMS evaluates every candidate across 10+ dimensions derived from your actual job description. Slide through 7 industries — the score, the compliance checks, and the underlying semantic matches all change together.
Same ten-plus dimensions, applied consistently across every industry — never a generic template, always derived from the real job description.
Skills & Competencies
Language Proficiency
Experience & Role Similarity
Salary Expectations
Industry & Domain Match
Seniority Level
Education & Certifications
Cultural Fit (AI-Driven)
Location & Availability
Career Growth Potential
Michael J.
Senior Full Stack Developer
London, United Kingdom
96%EXCELLENT
Score breakdown
Deterministic compliance
HOW SEMANTIC MATCHING WORKS FOR IT
We compare meaning, not words.
Every CV excerpt and job requirement is converted into a vector — thousands of numbers capturing meaning. The engine measures the angle between vectors: cosine similarity. Close angle, close meaning, regardless of shared vocabulary.
BEYOND THE SCORE — THE SAME SANDBOX, WHOLE FUNNEL
Screening is step one. Assess, interview, decide — in one place.
The platform also generates role-specific assessments from your job description, runs async video interviews, and keeps every review on the candidate's record. Try the demos below — sample data, fully interactive.
1 · A REST endpoint suddenly returns stale data after a deploy. Which is the most likely first thing to check?
Cache headers / CDN invalidation on the changed responses
Rename the endpoint and redeploy
Increase the server's CPU allocation
2 · A query filtering on customer_email is slow on a 10M-row table. The standard fix?
Load the table into application memory
Switch the column to full-text search
Add an index on customer_email
3 · Two services need the same validation logic. Best practice?
Copy the code into both services
Extract it into a shared, versioned library
Route every call through one service to validate
In the product, questions are generated from your job description (5–50, mixed formats). Multiple-choice auto-scores at submit; open and code answers go to AI review with per-answer human override — a score never auto-rejects anyone.
1 · A stakeholder demands a new feature mid-sprint and says it's urgent. What do you do first?
Add it quietly and work late to fit it in
Clarify the impact, then re-prioritise openly with the product owner
Decline — the sprint is locked, no exceptions
2 · You strongly disagree with a colleague's review comment on your work. Your next step?
Discuss it 1-to-1 with evidence, and escalate only if you still can't agree
Apply the change anyway to avoid conflict
Reply publicly explaining why they're wrong
Scenario answers in the product are reviewed against a hidden guidance rubric — AI proposes a score with rationale, and a human recruiter confirms or overrides every point.
Real sample answers — voice and all. Candidates record structured answers on their own time, no scheduling needed; recordings live on the candidate's application, gated to your hiring team.
Emily J.
Q1 · "Walk us through a project you're proud of" · 0:08
Hi! I'm Expertini's AI Product Expert. Ask me anything about our solutions, get guidance on any of our Hiring Tools, or just tell me what you're trying to do — I'll point you in the right direction. For account-specific issues, email support@expertini.com.
Thinking
Add your name and work email to start chatting — we use them only to follow up on your question if needed.
