Compliance/AI System Card

AI system card — TalenIA

System version: 2.2.x · Last updated: June 2026

This document provides public information about TalenIA's artificial intelligence system in line with the transparency principles of the Regulation (EU) 2024/1689 (EU AI Act). It is aimed at companies evaluating TalenIA for internal use, compliance officers, DPOs and candidates who want to understand how the system works.

System identification

System name

TalenIA — Candidate evaluation engine

Provider

Oriol Niella Hernandez · Tax ID 52917424J · Barcelona

EU AI Act category

High-risk system — Annex III, §4

Scope of application

Employment, workers' management and access to employment

AI models (primary)

Claude Haiku 4.5 — Anthropic PBC (USA)

Supported languages

Spanish · English · French · German

High-risk classification: AI systems used in personnel selection and evaluation processes fall under Annex III of the EU AI Act. TalenIA acknowledges this classification and has designed the product to meet its principles from the start, without waiting for the mandatory date.

What does this system do?

Inputs

  • 📄
    CVs in PDF format: Uploaded by the recruiter. The text is extracted on the server and the original file is discarded after processing.
  • 📋
    Job description (JD): Free text entered by the recruiter. May include requirements, responsibilities and organisational context.
  • 📊
    Psychometric responses (optional): 30 Likert questions (1–5) completed voluntarily by the candidate. Raw responses are deleted immediately after the report is generated.

Outputs

  • 🎯
    Fit scores: deliveryScore (0–10): fit of the delivery track record. contextScore (0–10): fit with the organisational environment. Both tied to the specific role, not to the candidate in the abstract.
  • 📝
    Qualitative assessment: Strengths (max. 3), risks (max. 3) and a natural-language summary. Always referenced to the job description.
  • 🏆
    Candidate ranking: Ordered by fit with the role. The recruiter can manually reorder it at any time.
  • Interview questions (optional): Segmented by type (validation / risk / clarification) and personalised per candidate and role.
  • 🧠
    Workplace Insights psychometric report (optional): 10 workplace-profile dimensions (0–100). Requires the candidate to voluntarily complete the test.
  • 📑
    Final combined report (optional): Combines CV evaluation with the psychometric report. Includes a final recommendation (hire / consider / reject) and key interview questions.

What this system does NOT do

  • It does not hire, reject or contact candidates autonomously.
  • It does not access social media or external data sources about candidates.
  • It does not process data under Article 9 GDPR (ethnic origin, gender, health, religion, sexual orientation).
  • It does not learn from or retrain on user data.
  • It does not generate credit scores or evaluations outside an active hiring process.

AI models used

ProviderModelUseTransfer
Anthropic PBCclaude-haiku-4-5Primary engine (paying users)USA — SCCs
Amazon Web Servicesclaude-3-5-haiku (Bedrock)Configurable alternativeUSA — SCCs
Google LLCgemini-2.5-flashFree tier (no prior purchase)USA — SCCs

SCCs = Standard Contractual Clauses (EU Implementing Decision 2021/914). TalenIA does not train its own models, it uses pre-trained foundation models exclusively via API.

Compliance measures implemented

Human oversight by design

No action affecting a candidate (advancing, rejecting, contacting) runs without an explicit recruiter decision. The system cannot act autonomously.

No special-category data

The scoring system does not process or infer data under Art. 9 GDPR. Scoring relies exclusively on declared experience and fit for the role.

Candidate pseudonymisation

The candidate's real name never appears in the recruiter interface. An automatically generated initials code is used instead.

PII purge (3 phases)

CV text and identifying data are irreversibly deleted at the end of the archive period (max. 60 days). Only numeric scores without identifiers are kept.

Explainability (XAI)

Every score includes a natural-language justification. The recruiter can read the system's reasoning and challenge it.

Uncertainty indicator

The system alerts the recruiter when CV data quality is thin or ambiguous (uncertaintyLevel: low / medium / high).

Transparency on rejection

When closing a process, the recruiter can send rejection emails with a template that mentions the use of AI. Transparency is the default.

Fallback on AI failures

If the model fails, the system activates deterministic engines that guarantee results without blocking the user or exposing sensitive data.

Name anonymisation in evaluation

The candidate's real name is never sent to the AI model. All evaluation engines (fit scoring, comparison, interview lens, final report) receive only the initials code generated when the CV is uploaded.

EU AI Office registration

Registration process started in the European Commission AI Office's public database (Art. 71 AI Act). Original deadline postponed from August 2026 to December 2027 (Digital Omnibus package); registration confirmation pending.

Known system limitations

  • The system evaluates only what is in the CV. It does not verify the accuracy of information declared by the candidate and has no access to external sources.
  • Non-deterministic outputs. Language models can produce slightly different evaluations of the same candidate across different runs.
  • Quality varies by language. Evaluation accuracy may be lower in languages with less representation in the foundation models' training data.
  • Optimised for senior and management roles. Entry-level roles (junior, internships) or highly specialised ones may produce less nuanced evaluations. Check the uncertaintyLevel field as an indicator.
  • Implicit bias inherited from the model. LLMs can reproduce patterns from their training data. Although the system minimises this risk through quantitative, role-linked criteria and no protected categories, implicit bias cannot be fully eliminated. Every hiring decision must be reviewed with human judgment.

Candidate rights

If you were evaluated using TalenIA in a hiring process, Article 86 of the EU AI Act grants you the right to:

  • Obtain an explanation of the results generated by the system and the factors that influenced the evaluation.
  • Request human review of the evaluation, without the AI score being the sole determining criterion.

To exercise these rights, contact the company that ran the hiring process. If you get no response, you can reach us at [email protected] stating the company and the role.

Full technical documentation

We maintain complete internal technical documentation under Art. 11 and Annex IV of the EU AI Act, including the Fundamental Rights Impact Assessment (FRIA) and a detailed architectural description of the system.

This documentation is available on request for:

  • The AESIA and other national supervisory authorities.
  • Client companies that need documentation for vendor risk, internal audit or DPIA processes.
  • Data protection officers (DPOs) evaluating the use of TalenIA within their organisations.