AI Disclosure — Witram LëtzDrill

Draft — not yet in force

For review by Luxembourg counsel before publication. Not legal advice.

Version 0.1, dated .

This page explains how and where Witram LëtzDrill uses artificial intelligence, what data is shared with the AI provider, how AI output is labelled, and how to turn AI off. It supports our transparency duties under the EU AI Act (Art. 50) and Apple's App Review Guideline 5.1.2(i).

1. In short

  • AI features are optional. The app works fully without them.
  • They use Google Gemini on Google Cloud, processed in the EU, called from our own server.
  • They are off until you give explicit consent in the app, which names Google Gemini.
  • AI output is labelled as AI-generated and can be wrong.
  • We do not use AI for emotion recognition, voiceprinting or biometric identification.
  • You can turn AI off at any time in Settings.

2. Which features use AI

FeatureWhat the AI does
Speaking and interview gradingTranscribes your spoken answer and gives a practice grade with feedback
Writing feedbackComments on your typed answer and suggests corrections
Mistake explanationsExplains why an answer was wrong
MnemonicsSuggests memory aids for vocabulary
Example sentencesGenerates extra example sentences using a word

Everything else — the curriculum, vocabulary lists, spaced-repetition scheduling, mock exam scoring of fixed-answer tasks, readiness and pace checks — works without AI, using rules and content written by people.

3. Provider

The AI model is Google Gemini, provided by Google Cloud (Gemini Enterprise Agent Platform) in the EU multi-region, so requests are processed in the European Union. Our server sends the request; your device does not contact Google Gemini directly.

4. What is sent

For each AI request we send:

  • your voice recording or typed text, and
  • the exercise prompt (the question or task).

We do not send your name, email address or account details.

Under the Google Cloud terms, Google does not use this data to train or fine-tune its models. We have turned off Google's response caching and don't use web search grounding.

5. What we keep

  • Voice recordings: deleted right after grading, by default. If you turn on "Keep my recordings", they are stored for 30 days so you can listen back, then deleted automatically.
  • AI grades and transcripts: kept with your learning progress until you delete your account.
  • Usage counters: kept for 90 days, for fair-use limits.

Details: Privacy Policy, https://letzdrill.witram.lu/legal/privacy.

6. How AI output is labelled

Every piece of AI-generated content in the app carries a visible "AI" label (for example "AI feedback" or "AI-generated example"). Content written or reviewed by people does not carry the label. [Where AI-generated Luxembourgish has been checked by a native speaker, it is marked as reviewed — confirm UI behaviour.]

7. Limits and accuracy

AI can make mistakes. In particular:

  • Luxembourgish is a smaller language, and AI models are less accurate in it than in English, French or German.
  • Speech transcription can mishear words, especially with background noise, accents or quiet recordings.
  • AI grades are practice feedback, not official scores. They are not an assessment by the Sproochentest examiners and do not predict your official result.
  • AI-generated example sentences or mnemonics may be unnatural, outdated or, rarely, incorrect.

If something looks wrong, trust the human-written lesson content and report it to support@witram.lu.

8. Human oversight

  • AI-written Luxembourgish that becomes part of the app's fixed content (for example example sentences we add to lessons) is reviewed by a native speaker before publication.
  • AI grading rubrics are designed by us based on the published exam format, [and we spot-check AI grades against human judgement — confirm this process exists before publishing].
  • AI output never makes decisions with legal or similarly significant effects on you (for example, it does not decide access to purchases or the Pass-or-extend guarantee).

9. What we do not do

  • No emotion recognition from your voice or text.
  • No biometric identification or voiceprinting.
  • No synthetic voices: all audio in the app is recorded by real people. We do not clone voices.
  • No profiling for advertising.

10. How to turn AI on or off

  • Turn on: when you first open an AI feature, the app asks for your consent and names Google Gemini as the provider.
  • Turn off: Settings → AI features → off. This takes effect immediately; AI features become unavailable and no further data is sent to Google Gemini.
  • Delete data: Settings → Delete account removes all stored recordings, grades, transcripts and usage logs.

11. Contact

Witram Tech S.à r.l.-S · privacy@witram.lu · support@witram.lu