Konzul is an AI-powered Hungarian citizenship interview simulator designed to recreate the unpredictability of a real consular conversation rather than simply test memorised answers.
Visit Konzul
Preparing for a naturalization interview is fundamentally different from preparing for a language exam. Applicants have to understand unpredictable questions, answer naturally in Hungarian, respond to follow-up questions and maintain a conversation under pressure.
Traditional vocabulary exercises, flashcards and static lists of interview questions do not recreate that experience.
The product shouldn't teach the interview. It should behave like the interviewer.
The central product decision was to treat the AI as a simulation environment rather than a tutor. A realistic consul cannot simply draw questions randomly from a fixed database. The system needs to understand the conversation, react to what the applicant says and decide what a plausible interviewer would ask next.
The interaction is designed as one continuous interview rather than a sequence of isolated language questions. The applicant enters a conversation that has a beginning, a direction and a momentum of its own, which is closer to what actually happens in front of a consul.
Follow-up questions depend on the conversational context instead of walking down a predetermined question tree. An incomplete or vague answer can be probed further, and a mentioned detail can become the subject of the next question.
Evaluation and feedback are delivered once the simulation ends, so the applicant can understand weaknesses without the live interview turning into a language lesson while it is happening.
A rigid interview tree would eventually become learnable. Contextual follow-ups make repeated sessions less predictable.
Less control over the exact path of any single session, in exchange for staying closer to the actual experience being prepared for.
General-purpose AI tends to help users. During a simulation, too much assistance undermines the purpose of the product.
The consul behaviour prioritizes interview realism over being a friendly tutor, which makes the experience deliberately less comfortable.
The simulation benefits from knowing enough about the applicant to ask plausible questions.
The product avoids storing unnecessary sensitive conversational data, accepting less personalisation than would technically be possible.
Correction during every answer would turn the experience into a lesson rather than a simulation.
Evaluation is separated from the interview itself, so useful corrections arrive later rather than in the moment.
The interview interface is only the surface. Underneath it, the conversation is held as state so the AI consul is answering to a conversation rather than to an isolated message, and the behaviour it follows is deliberately constrained rather than left to the model's defaults.
The result is a simulation with designed behaviour and product logic, instead of a chat window placed in front of a language model.
General AI behaviour naturally tends toward being cooperative, explanatory and helpful.
A consul simulation becomes less useful if the interviewer continually assists the applicant.
The interview behaviour and prompting were designed to keep the simulation focused on realistic questioning, contextual follow-ups and appropriate pressure rather than tutoring during the interview.
Konzul is now a publicly available product with the core experience running end to end: account creation, interview simulation, adaptive conversation, feedback and payment.
A powerful model does not automatically produce a convincing product experience. Behaviour, context and constraints matter enormously.
The model's natural tendency to assist can conflict with the purpose of a simulation product.
Collecting less information is preferable when the product can achieve its objective without retaining unnecessary personal data.