AI Transparency Notice

1. Why we publish this notice

Article 50 of Regulation (EU) 2024/1689 (the AI Act) requires people to be informed when they interact with an AI system, and artificially generated content to be marked. We also publish the conditions our model providers pass down to end users.

On this platform, KELVADO, S.L.U. is a provider and deployer of an AI system; we do not train foundation models of our own.

2. Where AI is involved

Feature	What the AI does	Who decides the result
Course generation	Turns documents and topics into modules, lessons, slides and quizzes	The author reviews and publishes
Chat assistant	Interprets natural-language instructions and edits the course	The author confirms each material change
Translation	Translates courses and materials into other languages	The author reviews the translated version
Images	Generates illustrations and covers	The author accepts or replaces them
Voiceover	Synthesises speech from the text with preset voices or, where the organisation enables it, with a voice cloned from a reference sample	The author approves the audio track
Learner tutor	Answers questions based on the course materials	Guidance only; it does not replace the instructor

3. Models and providers

We use third-party models through their APIs. The primary provider is Mistral AI (France): it generates and translates text, computes the embeddings used for search and, in the languages supported by its Voxtral speech model, synthesises the voiceovers. Voice cloning also runs on Mistral: the reference sample uploaded by the organisation is sent to Mistral to create the voice, Mistral retains it for a limited period (currently around thirty days) and, once it expires, we recreate it from the sample we store ourselves. For languages Voxtral does not cover (for example Russian), speech is synthesised with Azure AI Speech (Microsoft) in a European region. Images are generated with Replicate. The full, current list with countries and transfer safeguards is in Sub-processors.

4. What we do not do with your data

• We do not train models on your content, and we do not authorise our providers to do so.
• We send models only what is needed to produce the requested output.
• We use zero- or minimal-retention settings where the provider offers them.

5. Notice that you are talking to an AI

The chat assistant and the learner tutor identify themselves as artificial-intelligence systems in their first reply and in the conversation header. No automatically generated message is presented as written by a human.

6. Marking generated content

Content created with AI assistance is marked in the editor with an indication that it was generated and by which model. The organisation can decide whether that indication is also shown to the learner and in exports.

We are rolling out machine-readable marking (C2PA/IPTC metadata in images, metadata in audio and in exported packages) under article 50(2) of Regulation (EU) 2024/1689, which applies from 2 August 2026. This page will be updated as the roll-out completes.

7. Limitations and errors

Language models can produce incorrect, outdated or invented information. The output is a draft: it must be reviewed before publication and must not be used as medical, legal or financial advice. Human review is mandatory before publishing a course (clause 4 of the Terms).

8. Assessing people: a human decides

The platform can propose questions, mark multiple-choice answers and compute progress, but the final grade, the issuing of certificates and any employment or academic consequence are decided and validated by a person (instructor or training manager). No automated decisions producing legal or similarly significant effects on learners are made.

This design underpins our position on article 6(3) of the AI Act, documented internally.

9. A person's voice and likeness

Voice cloning requires the express consent of the voice owner, which the customer must retain and be able to evidence. Generating deepfakes or using a person's voice or likeness without authorisation is prohibited (Terms, clauses 4.1 and 4.2).

10. Contact

Questions about the use of AI on the platform or requests for further information: support@kelbado.eu.