AI Transparency

Last updated: October 9, 2026

Rigel’s promise: every match score can be traced to specific inputs you can inspect. This page explains where we use AI, where we don’t, and what matching means for you.

In short

AI reads CVs; rules score people. A language model helps turn your CV into a structured profile. Scores are calculated by fixed, published rules, with one exception described in Section 1.4.
You check the profile first. Nothing enters matching until you review and confirm it.
Every score can be explained, section by section, and you can switch matching off at any time.

1. Where we use AI

1.1 Reading your CV

CVs come in every format, so we combine fixed rules (text and layout analysis) with a large language model for the ambiguous parts: telling a job title from a company name, reading date ranges, recognising skill variants.

We use Google Gemini 2.5 Flash on Google Cloud Vertex AI. Your CV is processed within Google Cloud and, under Google Cloud’s terms, isn’t used to train Google’s models.

After parsing, you see the full draft profile and can edit every field before confirming.

1.2 Job descriptions (recruiters)

A recruiter can paste an existing job ad, and a language model fills the job form from it (title, skills, experience and education requirements). A recruiter can also ask Rigel to write a job description from the structured requirements they entered. Either way, the recruiter reviews and edits the result before publishing. Matching uses the structured requirements, not the description text. Both use the same Gemini model on Vertex AI.

1.3 Skill names

When you or a recruiter type a skill, we map it to a standard name (for example “React.js” → “React”). Fixed matching comes first; a language model helps only when a skill isn’t in our list yet, and the names it adds are recorded so they can be reviewed.

1.4 Relevant experience described in the recruiter’s own words

A recruiter can describe the relevant experience for a job in their own words (for example “B2B sales”) instead of picking from our list. For those jobs only, AI decides which of your past roles count towards your relevant experience:

an embedding model (Gemini embeddings on Vertex AI) compares each role in your profile with the recruiter’s words, and filters out roles that clearly don’t match;
a language model then judges each remaining role as fully, partly or not relevant.

The years in the roles that count feed the experience part of your score, with the same rules as any other job. We save each judgement, so the same profile gets the same result every time matching runs for that job. A role is judged again only if its text, the recruiter’s words, the model or our instructions to it change. The saved judgement is stored against a one-way fingerprint of the role and the recruiter’s words, with no name or ID. We record which roles were counted, so you can ask us: email privacy@rigelsignal.com, and a person will tell you and review any role you think was judged wrongly.

2. How matching affects you

This is profiling, and you should know exactly what it does.

You’re scored automatically. When you confirm your profile, it’s scored against every published job. When a recruiter publishes a job, it’s scored against every candidate with matching turned on. You don’t apply.
There’s a bar. Your score must reach 60 out of 100 for an application to be created for you automatically. A recruiter can raise the bar for a job, but not lower it.
Below the bar, that recruiter doesn’t see you for that job. That’s the main consequence of a score.
Above the bar, you’re anonymous. The recruiter sees your skills, experience and education, including employer names, but not your name, your contact details or the names of your schools: for education they see your degree and where you studied (the country, or the US state). You see the job title and your score. If a recruiter messages you before shortlisting, the conversation shows the job title, not the company.
A person makes the next decision. The recruiter reviews the ranked list and decides whom to shortlist. At shortlist, you see the full job and company, the recruiter sees your name and your school names, and you can decline that job.
Rigel never makes the hiring decision. Recruiters decide who to interview and hire.

Your choices:

See how a score was built: you see the skills, experience and education scores for jobs you’re shortlisted for, and your average breakdown in Insights. For the full breakdown of any score, email privacy@rigelsignal.com.
Correct your profile: edit it at any time; your scores are recalculated when you confirm the changes.
Stop profiling: turn off “Participate in matching” in Settings. Your profile then isn’t scored against any job. This is your right to object under the GDPR.
Decline a job you’ve been shortlisted for, with “Decline this match”.
Ask for a human review: email privacy@rigelsignal.com to have a person at Rigel check a score, give your view and contest it.
Export or delete your data: “Download my data” and “Delete account” are both in Settings. Details in the Privacy Policy.

3. Where we don’t use AI

3.1 Scoring

The matching engine applies fixed rules. A score is built from:

Skills: your confirmed skills against the job’s required and preferred skills, including level (beginner, intermediate, advanced, expert).
Experience: years of relevant experience against the requirement, with explicit bonus and penalty rules.
Education: degree level and field against the requirement.

Each section scores 0–100. The overall score is a weighted average using the weights the recruiter set (for example skills 50%, experience 35%, education 15%). No language model or embedding model calculates a score. The one place AI feeds into it is the judgement of which roles count as relevant experience for jobs described in the recruiter’s own words (Section 1.4), and because we save those judgements, the same inputs always give the same score for every job.

3.2 Ranking

Candidates are ranked by overall score, highest first. No engagement signals, personalisation or hidden re-ranking.

3.3 Bias

Rigel doesn’t evaluate “culture fit”, personality, or anything outside the job’s structured requirements. Scoring never uses names, photos or demographic data, and recruiters see candidates anonymously until shortlist. Structured scoring reduces bias; it can’t remove bias that is built into a job’s own requirements, which is why recruiters remain responsible for them.

4. Data used by AI

Inputs: the CV you upload; for recruiters, the job data or job ad they provide; for jobs described in the recruiter’s own words, the roles in your confirmed profile.
Outputs: a structured profile, job-form fields or description text, and relevance judgements for experience.
No training: we don’t use your data to train third-party AI models, or sell or share it for that purpose. All the models we use run on Google Cloud Vertex AI.
Logs: we don’t keep a separate log of what is sent to AI models. Our application logs are access-controlled, deleted after 30 days, and designed not to contain the content of your CV or of a model’s answer.

5. Why we built it this way

Many AI recruiting tools let a model read a CV and a job description and output a number. That score can change from day to day, and nobody can explain it. When a candidate asks “why not me?” or a regulator asks “how do you prevent discrimination?”, “the model said so” isn’t an answer — and increasingly isn’t a legal one.

Rigel uses AI for reading and rules for scoring: you get automatic profiles without the black box, and where AI does judge something, we say so here.

6. Regulation

Recruiters are responsible for their hiring decisions under the laws that apply to them, such as anti-discrimination and automated-hiring rules (for example New York City Local Law 144). We build Rigel to support those obligations and those that apply to us as the provider of the Service, including the GDPR and, as its requirements take effect, the EU AI Act.

7. Changes

If we change how we use AI — a new use, a different model, or different data practices — we’ll update this page and email active users.

Questions: privacy@rigelsignal.com