1. Why this policy exists

TalentyGo uses artificial intelligence to analyze resumes and suggest relevant job listings. In compliance with applicable AI regulations including the EU AI Act 2024 (Regulation 2024/1689) Article 50 (transparency obligations) for EU users, we publish this policy to explain how our system works in accessible terms.

AI-based matching for recruitment is classified as a high-risk AI system under Annex III paragraph 4 of the EU AI Act (for EU users). This means we are subject to enhanced transparency, audit, and governance obligations.

2. What Charlie AI and AVA are

2.1 Charlie AI

Charlie AI is TalentyGo's analysis and matching engine. It comprises:

Resume skill extraction pipeline (multi-format parser: PDF, DOCX, TXT)
Large Language Models (LLMs) for semantic parsing and classification
Vector embeddings for semantic matching between user profile and listings
Ranking system assigning a "match score" from 0 to 10 to each listing

2.2 AVA — AI Career Coach

AVA is TalentyGo's conversational interface. A 29-year-old AI character designed to provide career suggestions, generate cover letters, and conduct mock interviews. AVA is not a real person.

3. Technology used

TalentyGo uses the following third-party AI components:

LLM models: GPT family (OpenAI) and Claude (Anthropic) for resume parsing, cover letter generation, interview simulations.
Embeddings: text-embedding-3-large (OpenAI) for semantic resume-listing matching.
Vector store: Cloudflare Vectorize for fast semantic search.
Pre-processing: proprietary skill extraction + entity recognition pipeline.

All data sent to LLM providers is pseudonymized before transmission (we remove email, phone, address, personal identifiers not strictly necessary for matching). Providers cannot identify individual users.

We have Data Processing Agreements with all providers, including a prohibition on training on customer data (training opt-out).

4. What Charlie AI does

Skill extraction: identifies technical skills, soft skills, languages, frameworks, certifications from your resume.
Seniority inference: estimates experience level (Junior, Mid, Senior, Lead) based on years and roles.
Preferences inference: extracts cities of interest, employment type sought, preferred industry.
Semantic matching: compares your profile with aggregated listings by computing embedding cosine similarity.
Ranking: orders listings by relevance assigning a transparent "match score" (visible to user for each listing).
Generation: produces cover letters, resume optimization, interview simulations on your explicit request.

5. What Charlie AI does NOT do

Does not make hiring decisions. Decisions are human: employers decide who to hire, users decide where to apply.
Does not evaluate users discriminatorily. The system is trained to avoid bias on gender, ethnicity, religion, orientation, nationality, age. We conduct periodic audits on a synthetic dataset to verify absence of systematic biases.
Does not profile based on special categories under Art. 9 GDPR. We do not use as matching features any data on health, religion, political opinions, sexual orientation, ethnicity.
Does not share data with third parties without consent. LLM providers receive pseudonymized data for processing only.
Is not trained on user data. The LLM providers we work with are configured in "no training" mode (zero data retention for training).
Does not manipulate results to favor paying employers. TalentyGo does not receive compensation from employers to "push" listings to users. Ranking is based exclusively on relevance computed by matching.

6. Your rights regarding AI

You have the following specific rights:

Right to know: you know you are interacting with an AI system. All content generated by Charlie AI / AVA is explicitly labeled as AI-generated.
Right to opt out: you can disable AI matching and use TalentyGo as a traditional search engine. To disable AI matching, email [email protected].
Right to human intervention: you can request that an AI decision be reviewed by a human operator by emailing [email protected].
Right to express your point of view: you can contest the match score of a specific listing and provide feedback to improve the system.
Right to explanation: for each listing in the roster you can click "Why this match?" to see concrete criteria (e.g., "skill match 8/10, location Milan, seniority compatible, industry SaaS").
Right to deletion: you can request deletion of your data (including AI derivations like resume embeddings).

7. Governance and audit

For EU users, in compliance with high-risk AI system requirements (Arts. 8-15 EU AI Act), we implement:

Risk management system: periodic risk assessment of the AI system (Art. 9).
Data governance: training data quality, bias mitigation, validation set (Art. 10).
Technical documentation: kept up-to-date (Art. 11).
Record-keeping (logging): AI decision logs retained for 24 months for audit (Art. 12).
Transparency: this policy, links on landings, "explain decision" option in product (Art. 13).
Human oversight: mechanisms enabling human intervention (Art. 14).
Accuracy + robustness: matching accuracy metrics monitored, fallback in case of failure (Art. 15).

8. System limitations

Charlie AI is a powerful but not infallible system. Known limitations:

Does not always correctly extract skills from non-standard format resumes, images, or rare languages.
Match score is a statistical estimate, not a guarantee of real fit.
Aggregated listings may already be closed at the time of match.
System may not recognize recently-emerged skills (temporal update gap).
AI reflects patterns present in training data: residual undetected biases may exist.

If you find inappropriate behavior (e.g., discrimination, systematic errors), report it to [email protected].

9. AI Governance contacts

General AI: [email protected]
AI decision review: [email protected]
Bias or error feedback: [email protected]
DPO (Data Protection Officer): [email protected]