CrewHallCrewHall
AI Transparency — Art. 13 EU AI Act (EU) 2024/1689

How We Use Artificial Intelligence

CrewHall uses AI systems to facilitate recruitment. This page explains in detail what AI we use, how it works and how it affects your experience, in accordance with Art. 13 EU AI Act obligations.

High-Risk AI System — Annex III, Point 4 EU AI Act

CrewHall is classified as a high-risk AI system because it is used in the career matching and job placement process. This means we apply the highest standards of transparency, human oversight and protection of your rights.

1. AI Career Coach

ElevenLabs
What it is:A conversational AI career coach that builds your career profile through a voice session.
Provider:ElevenLabs Inc. (USA)
How it works:Your voice is transmitted in real-time to ElevenLabs servers, transcribed and processed by an AI agent that extracts information about your profile (position, experience, skills, salary).
Data processed:Voice audio, text transcription, profile information provided verbally.
Storage:Transcriptions and extracted profile data are saved in the CrewHall database (Supabase, EU). Raw audio data is managed according to ElevenLabs policy.
Consent:Mandatory consent is requested before every career session (Art. 50 EU AI Act).
Right to refuse:You can quit the session at any time. You can also fill in your profile yourself, from your dashboard, without the voice interview.

Important: The AI coach does not make hiring decisions. It collects information for your career profile. The final decision belongs exclusively to the human employer.

2. AI Competency Assessment

Google Gemini
What it does:After a voice session, an AI model reads the transcript of what you said and produces an indicative score (0–100) for each competency defined for the role, together with a written rationale for each score.
Provider:Google — Gemini 2.5 Flash (USA)
Data processed:The text transcript of your voice session. This is personal data: it contains what you said about your experience, your work history and your skills. Raw audio is not sent to the scoring model.
What is stored:For every score we store the value, the written rationale, the weight applied, the exact model identifier and the prompt version used — so any score can be reconstructed and explained later (Art. 12 EU AI Act).
Impact:Scores are indicative and are shown to employers as one signal among others. They do not automatically reject you and they do not by themselves decide whether you are hired.
Known limitations:The model scores what was said in one conversation. A short session, a bad connection, an unfamiliar accent or nervousness can lower a score without reflecting your actual competence.

You can contest a score. If you believe a score misrepresents you, you can record a contestation from Settings. Your request is recorded with the reason you choose and the time you sent it.

3. AI Matching Algorithm

High-Risk
What it does:Calculates a compatibility score (0–99) between a candidate and a job.
Technical model:Proprietary algorithm (matching.ts) — no external ML. Runs locally in the browser.
How the score is calculated:
44% Skills overlap (keyword matching)
30% Title & skills similarity (Jaccard)
26% Semantic similarity (cosine)
+4 Professional domain bonus (if matched)
Domains:The algorithm automatically classifies profiles into: trades, maritime, culinary, it, security_it, engineering, medical, finance, legal, education, other.
Known limitations:The score is less accurate for: profiles with fewer than 3 skills, ambiguous titles, interdisciplinary domains, or profiles with very short text.

Human oversight (Art. 14 EU AI Act, our obligation): The AI score is advisory only. Employers can ignore the AI ranking and evaluate candidates manually. You can contest your matching score from Settings.

4. Automatic Domain Classification

Google Gemini
What it does:Automatically classifies each candidate profile and each job into a professional domain.
AI model:Google Gemini 2.5 Flash — runs exclusively on the server.
Data processed:Job title, skills list, industry — no personally identifiable data.
Impact:The classified domain influences matching: candidates are matched with jobs from the same domain.
Fallback:If AI classification fails, the domain defaults to 'other' and matching is based on keywords only.
Security:The Gemini API key is stored exclusively on the server (Supabase Secrets). It is never sent to the browser.

5. Eligibility Checks

Automated decision
What it does:Checks whether you meet the mandatory requirements of a specific job: the certificates the role requires and, where applicable, the minimum days of sea service.
How it decides:These checks are NOT made by an AI model. They are fixed rules applied to the documents and sea service you entered: a certificate is present or missing, valid or expired; sea days meet the threshold or do not.
Result:An eligible / not eligible verdict for that job, with a written explanation of which requirement failed.
Why it matters:This verdict is produced automatically and can affect whether an employer considers you for a role. It is run on demand — when you or an employer asks for it — and every run is logged (Art. 12 EU AI Act).
What employers see:Employers receive the verdict and which requirements were met or missed. They do not receive your document files or your exact number of sea days.

Your right to human intervention (GDPR Art. 22(3)): a "not eligible" verdict is a decision made without a human. GDPR gives you the right to human intervention in such a decision — for example if a certificate is valid but was recorded incorrectly. Record a request from Settings.

6. What the AI Does NOT Do

  • ✓Does NOT make final job placement decisions — these belong exclusively to the human employer.
  • ✓Does NOT access or process sensitive data: ethnic origin, religion, health, political orientation.
  • ✓Does NOT analyse emotions or voice tone for psychological evaluation purposes.
  • ✓Does NOT use age, gender or nationality as scoring factors.
  • ✓Does NOT share your profile with third parties without your consent.
  • ✓Does NOT make irreversible automated decisions without the possibility of contestation.

7. Your Rights

GDPR + EU AI Act

Right to information (GDPR Art. 13)

This page.

Right to explanation (GDPR Art. 22)

Each competency is shown with the written reasoning behind its score.

Right to human intervention (GDPR Art. 22(3))

You can record a request about an automated verdict from Settings.

Right to contest (GDPR Art. 22(3) + AI Act Art. 86)

You can contest your matching score from within the app.

Right to erasure (GDPR Art. 17)

Settings → Delete Account → all your data is removed.

Right to portability (GDPR Art. 20)

Settings → Download My Data → full JSON export.

Human oversight (AI Act Art. 14 — our obligation)

Employers can evaluate manually, ignoring the AI score.

8. Contact & Complaints

To contest an automated matching score, use Settings → Contest My AI Matching Score in the app (GDPR Art. 22(3) + AI Act). For other questions about AI usage or GDPR rights:

Email:privacy@crewhall.co
Legal basis:EU AI Act (EU) 2024/1689 + GDPR (EU) 2016/679
Last updated:August 2026