The problem it solves
Most people prepare for an interview by reading lists of typical questions. What they rarely get is practice: saying the answers out loud, to someone who asks back, in the language the interview will be held in. A friend does not know the job, and a coach is not available at eleven at night before the big day.
What it does
The user uploads a job posting and their CV. From these two documents the service builds a personalised interview training around them. It covers every stage of a real interview, from the introduction to the salary discussion, and it is conducted by voice, in German.
It is a full product, not a demo: user accounts, personal dashboards and payments are part of it.
How it is built
At its core is a voice-first conversational AI that runs realistic practice sessions. The part I care about most sits under the conversation. Every line the AI says is checked against a live model of what the person already knows. That keeps the conversation at the right level and introduces one new thing at a time instead of burying the user in questions. Session history drives what comes next, so the second session does not repeat the first.
The same engine runs a second application, onboarding for new employees, which is described as its own case.
Privacy by design
A CV is personal data by definition, and a practice interview is a conversation people would not want to be heard by others. I design this kind of system the way I design every other: the AI receives the context it needs for the current step, and raw input is separated from the result. CVs and job postings stay with the user.
Where it can go next
The service was built for the candidate. The same engine can face the other way, towards the company that runs the interviews.
What it shows about how I build
A product like this is a scenario before it is technology: a person has an interview on Thursday and wants to be ready. Voice is the interface because an interview is spoken. The AI decides how the conversation continues, but it decides against a model of what the person already knows, not by improvising. And it is built as a product people pay for, with accounts and dashboards, not as a prototype.