How AI is used inside Konzul — and where it stops.

What the AI does

• Generates practice conversations. A language model phrases questions in natural Hungarian and follows up on what you said, inside a topic structure and stage plan authored by the Konzul team.
• Transcribes your speech. A speech-to-text model converts what you say into text so it can be assessed.
• Speaks aloud. A text-to-speech model produces the spoken Hungarian you hear.
• Writes language feedback. A model drafts observations about grammar, vocabulary range, fluency and topic coverage, and estimates a practice level.

What the AI does not do

• It does not make legal, immigration or administrative decisions of any kind.
• It does not represent any government, embassy, consulate or authority.
• It does not assess your eligibility for anything, or predict an official outcome.
• It does not review documents for official purposes.
• It does not give legal or professional advice.

Known limitations

AI systems make mistakes. Konzul's practice partner can mis-hear a word, mis-transcribe an accent, accept an answer that a human would question, correct something that was already right, or produce a statement of fact that is out of date or simply wrong. Feedback scores are estimates from a single practice session, not measurements.

Treat everything Konzul produces as practice material. Verify any factual claim, requirement, deadline or procedure using official embassy, consulate or government resources before relying on it.

Educational purpose

Konzul is intended for educational use: building spoken fluency, vocabulary and confidence in Hungarian. It is not a substitute for a teacher, a translator, a lawyer or an official information source, and progress inside Konzul does not guarantee any real-world result.

Data sent to AI providers

Konzul uses OpenAI models. Audio is streamed for transcription and discarded — it is never stored. Before text leaves our servers, identifiers such as document, tax or card numbers are stripped and replaced with neutral placeholders. Requests are stateless, provider-side retention and history are switched off, and your content is not used to train models. Full detail is in our Privacy Policy and Security page.

Human oversight

The Konzul team authors and reviews the learning structure, the topics, the feedback criteria and the safety rules the models must follow — including a rule that the practice partner must never ask for identification numbers, payment details or passwords. If you see AI output that looks wrong, unsafe or inappropriate, please report it so we can correct it.