Practise the interview before the pressure is real.
Rolebird builds mock interviews from your CV against that job - the questions your experience actually invites, follow-ups that react to what you say, and an honest debrief after every session. Typed, or out loud with your speech transcribed on your own device.
One pack covers one job. Nothing renews, no subscription, and no fixed expiry while your account stays active.
What one pack includes
15 practice sessions - more than almost anyone needs - plus one final mock, taken once, when you say you’re ready. The pack binds to one job the first time you use it, because interview preparation is specific: the questions come from that listing and your CV, not a generic bank.
$49.99 one-off - less if one of the first-100 founding slots is still open; the live counter is on the pricing page. A stated 15, never “unlimited”: the industry’s unlimited always hides a fair-use clause, and a printed number is the same offer with the limit on the outside.
How the sessions work
Planned from your evidence
Each session is planned from your CV and the job ad. A handful of core questions, each with room for a follow-up that reacts to what you actually said - the way an interviewer probes, not a script that ploughs on regardless.
It learns what to drill
After each session you get a debrief: what held up, what to work on. The next session is planned knowing what the earlier ones showed, so the thin spots get drilled instead of the answers you have already nailed.
Curveballs are earned
Your first session builds trust in the format. Repeat sessions each include one curveball - a question you will not have prepared for, drawn from your real evidence - because composure on the unprepped question is exactly what practice is for.
The questions you’d rather not be asked
If your CV shows an employment gap, a short tenure, a career change, or experience that’s all overseas, the interviewer will ask about it - because a real one will. Rolebird challenges the thinnest parts of your evidence and helps you explain them honestly, rather than teaching you to disguise them. Only questions your documents actually invite: nothing is drawn from a generic “tough questions” list.
The final mock
When you say you’re ready - and only after you’ve practised - you can take the final mock. It is longer than a practice session, runs the arc of a real interview from warm-up to “what questions do you have for us?”, includes two curveballs, and is weighted toward whatever your practice history says is still thin.
It is taken once, like the interview it rehearses. Afterwards you get a graduation brief: where you are strongest, what to review on the day, and your own best answers quoted back to you - because the answer you already gave well is the one worth remembering in the room.
What a session feels like
An illustrative exchange - written for this page, not taken from anyone’s session.
Interviewer: Your CV says you supervised weekend stock counts across two sites. Walk me through a count that went wrong.
You: One weekend the second site’s numbers came in double-scanned. I picked it up comparing the variance report against the previous month…
Interviewer: You caught it at review. What would have caught it two hours earlier?
Voice, and what it needs
Typing is the full product
Every session works fully typed - same questions, same follow-ups, same debrief. Voice is a mode, not a requirement.
Your voice stays on your device
In voice mode the interviewer speaks and you answer out loud. Your speech is transcribed on your own device - the audio is never uploaded and never stored; only the resulting text is submitted, exactly like a typed answer.
What voice mode requires
A desktop browser, microphone permission, and a one-time speech-model download - at least ~500 MB, and ~1.6 GB for the recommended model - into your browser before your first voice session. Spoken answers run up to two minutes each. On a phone, or without the download, type instead.
Two speech models - which and why
The interview transcribes your speech on your own device, so you choose the model that downloads. Both are one-time downloads your browser keeps; you can switch later.
Standard · ~500 MB
The lighter download, and it runs on more machines - no dedicated graphics needed. Best when you speak clearly at a steady pace.
English only, and it can stumble on fast, quiet or strongly accented speech, and on te reo Māori words and place names.
Best · ~1.6 GB - recommended
Larger and multilingual, so it keeps up with fast, quiet and accented speech, and recognises the te reo Māori kupu and place names that pepper New Zealand English - which the English-only model can garble. For most NZ users that accuracy is worth the bigger download, which is why we recommend it.
It wants a desktop browser with WebGPU and enough memory to hold ~1.6 GB; where that is not available, the Standard model is used instead.
Both transcribe entirely on your device - the model choice changes accuracy and download size, never where your voice goes. It never leaves the machine.
Will voice mode work on this device?
A quick check of this browser - it does not turn on your microphone or send anything.
What we keep, and what we never have
Session transcripts and debriefs are stored encrypted, belong to your account, appear in full in your data export, and are deleted when your account is. Your audio is never among them - it never left your device. The interviewer’s spoken questions are generated from the session’s own text; nothing you say is ever turned into audio.
What this is, and is not
This is practice: a structured way to rehearse defending your real experience before a real person asks about it. It is not a prediction of your interview, a guarantee of an offer, or a score - there are no scores, deliberately. A number would tell you how you compare to nobody in particular; the debrief tells you what to do next.
And it will not coach you to bluff. If an answer would surprise your referees, it has no business in the interview.
That evidence-grounding - the rule that every claim traces to something real - runs through the whole system. When we tried to attack it, it mostly held: our internal adversarial test.