Pick a round, answer out loud to an AI interviewer for twenty minutes, and read written feedback as soon as you finish. Sharpen your delivery on realistic scenarios and build confidence for every stage of the hiring process.
AI Interviews
Every interview above is a twenty minute spoken session with an AI interviewer, ending in written feedback. This guide explains exactly what happens from the moment you press join, what the feedback covers, how to prepare, how it differs from a mock interview with a person, and what we record while you talk.
An AI interview is a spoken practice interview run by software instead of a person. You pick an interview from the list above, your browser asks for microphone and camera permission, an AI interviewer named Cade asks a question out loud, and you answer out loud. Your answers are transcribed as you speak, and written feedback with improvement suggestions appears when the session ends.
The point is repetition under mild pressure. Most candidates who fail interviews are not short of knowledge, they are short of practice at saying what they know while a clock runs and someone waits. Reading a solution builds recognition. Saying it aloud, unprompted, from a blank start, builds the skill the interview actually measures, and that only comes from doing it.
Each interview in the list above carries three labels worth reading before you start: a category, a difficulty of Easy, Medium, or Hard, and a duration of twenty minutes. Some are marked Free and some cost one credit. Use the search box and the category filters to find the round type you want rather than starting with whatever is at the top.
Nothing here is a surprise once you have done it once, but the first session catches most people out on the noise check or the silence detection. This is the whole sequence in order.
Filter by category or search by title, then read the three labels on the card before you commit. The difficulty tells you how hard the questions will push, the twenty minute duration tells you the time budget, and the Free or one credit label tells you what it costs. Start one level below where you think you are.
Browsing the list is open to everyone, but starting a session needs an account, because your transcript and feedback have to belong to someone. If you are not signed in when you press join, a sign in prompt appears and the session continues once you are through it.
The pre join screen shows a camera preview and a status line for your camera and microphone. Press check system and your browser asks for permission. Both have to read ready before you can continue, and if your browser has blocked microphone access the page tells you exactly where to change it.
After both devices are ready, a short noise check runs. A bar shows your current input level and asks you to stay quiet for a few seconds. It has to settle to a low level before the join button becomes active. This exists because the interview decides you have finished speaking by detecting silence, and constant background noise breaks that.
Press join and the session connects. Cade, the AI interviewer, opens with a greeting and the first question, spoken out loud. Wait until it has finished talking before you start. There is no transcript on screen to read from, so treat it exactly like a phone screen and listen properly.
Speak your answer as you would to a human interviewer. Your microphone audio is recorded and transcribed while you talk. When you go quiet for around three seconds, that is the signal that you have finished, and your answer is submitted. Long silences in the middle of an answer will cut you off, so think out loud rather than pausing to think.
The interviewer replies to what you said and moves on. Because the session is twenty minutes and has to be completed in one sitting, treat the clock as part of the exercise. Answer in a couple of minutes and stop, rather than filling the time, because a rambling answer scores worse than a short complete one.
When the interview ends, an interview completed panel appears with written feedback and, where there is something specific to act on, improvement suggestions. Read it immediately, then write down the one thing you are going to change before your next session. Feedback you do not turn into a next action is entertainment.
At the end of a session you get written feedback on how you answered, plus improvement suggestions where there is something concrete to change. It is generated from the transcript of what you actually said, which means it is responding to your words rather than to your intentions, and that is precisely what makes it useful. Interviewers also only hear your words.
Read it with one question in mind: what did the transcript show that I did not think I was doing? The recurring findings are almost always structural rather than technical. Answers that never state a conclusion. Answers that start in the middle. Answers that list options without choosing one. Answers that go on for four minutes when ninety seconds would have been complete. Those are habits, and habits respond quickly to being named.
Be careful not to treat the feedback as a score to beat. Optimising for a generated assessment produces answers that sound like they were written for a rubric, which is its own failure mode in a real interview. Use it to spot patterns across several sessions instead. If three transcripts in a row show you never stated your assumptions, that is a finding worth acting on. One transcript is noise.
The other thing worth doing is reading your own transcript rather than only the summary. Almost everyone is surprised by how much filler is in their speech and how often they start a sentence twice. You cannot hear that live. You can see it immediately in text, and once you have seen it you start noticing it while you talk.
Preparation for an AI session is different from preparation for a study session, because you are rehearsing delivery rather than acquiring material. Half an hour of the right preparation makes a twenty minute session worth far more.
Sort out the environment first, since that is what wastes most people's first attempt. Find a room where you can talk out loud without interruption, use a headset if you have one, close the window, and turn off anything that hums. The noise check has to settle before you can join, and a laptop microphone in a room with a fan often will not settle at all.
Then prepare the content lightly. Have a two minute version of your own background ready, because a lot of interviews open there and fumbling the easiest question sets a bad tone for the next eighteen minutes. Pick the category you are going to practise and skim the relevant material for twenty minutes beforehand, not two hours, because the session is testing recall under pressure and cramming defeats the purpose.
Finally, decide in advance what you are practising. A session where you are working on being concise is a different session from one where you are working on stating assumptions, and trying to fix everything at once fixes nothing. Write your one focus on a piece of paper and put it next to the screen.
One thing not to do: do not write your answers out and read them. The transcript will be fluent and the exercise will be worthless, because reading aloud uses none of the machinery that fails you in a real interview.
These two formats are not competitors, they are different stages. AI practice is high volume, low friction, and available at midnight. A mock interview with an expert is low volume, high signal, and scheduled. Using only one of them is the mistake, and the more common version of that mistake is doing neither and reading instead.
What AI practice is good at: repetition. You can run a session tonight, another tomorrow, and a third at the weekend, which is how spoken fluency is built. It never gets bored of your third attempt at the same question, it never gives you the benefit of the doubt because you seem nice, and it produces a transcript, which is a more honest record of your answer than anyone's memory.
What it cannot do: read the room. A human interviewer notices that you hedged, pushes on the part you were hoping to skip, follows a promising tangent, and can tell you that your answer was technically correct and still would not have passed at their company. That judgement is why the expert mock interview packages exist, and it is why they include feedback with concrete improvement areas and a recording you can watch back.
The sequence that works for most people is to build fluency with AI sessions until the basic mechanics are automatic, then spend money on a human mock once you are no longer wasting the hour on nerves and structure. Booking an expert session while you still stumble over your own background is paying senior engineering time to watch you warm up.
Solo study belongs in the picture too, and it is where the material comes from. The failure pattern to watch for is a preparation plan that is ninety percent reading, because reading feels productive and never gets tested. If you have not said an answer out loud this week, your ratio is wrong regardless of how many problems you have read.
These three do different jobs and most preparation plans are badly balanced between them. The usual imbalance is far too much reading and almost no speaking.
| AI interview practice | Mock interview with an expert | Studying on your own | |
|---|---|---|---|
| What it builds | Fluency at saying answers out loud under a clock | Calibration against a real hiring bar, and a plan for what to fix | The underlying knowledge |
| Time per unit | Twenty minutes | Sixty minutes, booked in advance | As long as you give it |
| Feedback | Written feedback and improvement suggestions from your transcript, immediately | Feedback with concrete improvement areas, plus a recording of the session | None, unless you record yourself |
| Availability | Any time, as often as you like | Scheduled with a person | Any time |
| Cost | Free interviews, or one credit each | Paid packages | Free, or the price of a course |
| Main risk | Optimising your answers for a generated assessment | Booking it too early, before nerves and structure are handled | Feeling productive without ever being tested |
Practice works when it is scheduled and fails when it is spontaneous, because a spoken session takes a small amount of courage and there is always something easier to do instead. A workable week has three slots, and none of them is long.
Early in the week, one twenty minute session on your weakest round type, chosen deliberately rather than by mood. Read the feedback the same evening and write down the single change you are making. Mid week, one study block of an hour or two on whatever the feedback exposed, using a course or the interview questions library rather than random browsing. Late in the week, a second session on the same round type to test whether the change stuck.
Around that, two habits. Keep a running list of every question you were asked and how you answered it, because after a month that list is a better study guide than any general resource. And close the loop publicly: post what you learned in the community, because writing it down for someone else is the step that turns a session into knowledge you keep.
Once a month, or before a real loop, book a mock interview with an expert. That cadence means each paid session lands on a version of you that has changed since the last one, which is what makes the feedback worth what it costs. Booking two in the same week mostly gets you the same feedback twice.
The whole routine is about three hours a week. It beats twelve hours of reading, and the reason is not effort, it is that speaking is the skill being tested and reading is not.
Being straightforward about this matters, because you are being asked to turn on a microphone. When you take an AI interview we record your microphone audio for the duration of the session and generate a transcript from it, along with scores and written feedback. Your browser asks for microphone permission before any of that starts, and you can decline or stop at any time.
The transcription and the feedback are produced with the help of OpenAI, which receives the interview audio in order to transcribe it and to generate feedback. That is the only reason your audio leaves your machine, and it is listed in our privacy policy alongside every other provider we use and what each one receives.
Recordings, transcripts, and results are kept while they are relevant to your applications and to the employers you have shared them with. You can withdraw consent, including by revoking microphone access in your browser, and you can ask us to delete your data. The privacy policy sets out those rights and how to exercise them.
The camera is checked before you join, so a camera preview is part of the pre join screen and both devices have to read ready before the join button activates. If you are practising somewhere you would rather not be filmed, practise somewhere else rather than trying to work around the check, because the check is there to stop sessions failing halfway through.
One practical note that is easy to miss: because the interview decides you have finished speaking by detecting a few seconds of silence, your microphone is listening throughout the session, not only when you intend to answer. Treat the whole twenty minutes as recorded, and do not hold a side conversation in the room.
The two failure modes worth knowing about in advance are both mechanical, and both are fixable in a minute once you know what is happening.
The first is a session that keeps cutting you off. That is the silence detection doing its job: a pause of a few seconds is read as the end of your answer. The fix is to narrate instead of pausing. Say what you are about to do before you do it, and keep a low murmur of reasoning going rather than going quiet to think. This is a genuinely useful habit for real interviews too, because a silent candidate is an unreadable candidate.
The second is a session you cannot start, which is almost always the noise check. If the level bar will not settle, you are in a room with a constant sound source. Fans, air conditioning, traffic through an open window, and a mechanical keyboard are the usual suspects. A headset microphone solves it more reliably than anything else, because it hears you rather than the room. Also check you are in Chrome, since that is the supported browser.
If a session goes badly for content reasons rather than technical ones, do not immediately start another. Sit with the feedback, work the gap, and come back. Three sessions in a row on the same afternoon produce three transcripts of the same weakness. One session, one fix, one more session next time produces progress you can measure.
And if the interviewer misunderstands you, notice that this happens to real interviewers too. The correct response is the correct response in a real interview: notice the mismatch, say that you may have misunderstood the question, restate what you think is being asked, and continue. Practising that recovery is worth more than a smooth session.
Interviews marked Free on the cards above cost nothing. The rest use one credit from your wallet, and the pre join screen tells you before you start, with a link to check your balance.
Browsing the list, using the search box, and filtering by category are all free and need no account.
Yes. You can browse and filter the list without signing in, but starting a session needs an account, because your transcript, feedback, and credit balance all belong to you.
If you press join while signed out, a sign in prompt appears and the session continues once you are through it.
Twenty minutes, and it has to be completed in one sitting. Budget half an hour in total so you have time for the system check beforehand and the feedback afterwards.
There is no pause. If you leave the tab or take a call partway through, you lose the session.
Both. The pre join screen shows a camera preview and a status line for each device, and the join button stays disabled until both read ready and a short background noise check has settled.
Use Chrome. It is the supported browser, and other browsers may fail the media checks.
Yes. Your microphone audio is recorded for the duration of the session and turned into a transcript, along with scores and written feedback. Your browser asks for microphone permission before this starts and you can decline or stop at any time.
The audio is transcribed with the help of OpenAI, which is why it leaves your machine. Recordings, transcripts, and results are kept while they are relevant to your applications and to the employers you shared them with. The privacy policy names every provider we use and sets out how to request deletion.
Because it decides you have finished speaking by detecting a few seconds of silence. A long pause to think is read as the end of your answer.
The fix is to narrate rather than pause. Say what you are about to do before you do it and keep a low run of reasoning going. This is a good habit for real interviews too, because a silent candidate is an unreadable candidate.
The level bar has to settle to a low reading before the join button activates, and a constant sound source will stop it. Fans, air conditioning, an open window onto traffic, and a loud keyboard are the usual causes.
A headset microphone is the most reliable fix, because it picks up you rather than the room. Also close other tabs that might be playing audio and confirm you are in Chrome.
Written feedback on how you answered, generated from the transcript of what you actually said, plus improvement suggestions where there is something specific to change. It appears in a panel as soon as the interview ends.
Read it the same day and pick one change to make. Judge patterns across three or more sessions rather than reacting to a single one.
AI practice gives you unlimited repetition, an immediate transcript, and no scheduling, which is what builds spoken fluency. A mock interview with an expert gives you human judgement, calibration against a real hiring bar, and a recording you can watch back.
Do the AI sessions first until the mechanics are automatic, then book an expert mock. Paying for senior engineering time while you still stumble over your own background is paying someone to watch you warm up.
Two twenty minute sessions a week, with one study block between them aimed at whatever the first session exposed, is enough for most people. That is about three hours a week including the reading.
Running several sessions back to back on one afternoon mostly produces several transcripts of the same weakness. One session, one fix, then another session is what produces measurable change.
An Easy one in a category you already know well. The purpose of the first session is to learn the format, not to be assessed, and it is much easier to notice how the pacing and the silence detection work when the content is not also stretching you.
After that, choose by weakness rather than by mood. The round type you least want to practise is nearly always the one to book.
Yes, and doing so deliberately is one of the most effective things you can do. Repeating the same interview a few days after making one specific change is the clearest way to feel that the change worked.
Credit costs apply per session, so check whether the interview you want to repeat is a free one or a one credit one.
Yes, and behavioural rounds are arguably where spoken practice pays off most, because those answers live or die on structure and length rather than on knowledge. Filter the list by category to find the round type you want.
The transcript is especially useful here. Reading back a behavioural answer shows you immediately whether you actually said what the outcome was, which is the part most people leave out.
Practice tells you what is weak. These are the pages that fix it, and the ones that handle the rest of the search.
Sixty minute sessions with engineers and hiring managers, with feedback and a recording afterwards.
Questions by role and company, so you can choose what to practise instead of guessing.
Ask people who sat your target loop what it contained, then practise exactly those rounds.
What we record during an interview, who processes it, how long it is kept, and how to have it deleted.
Systematic coverage of what algorithm rounds test, built for people starting from scratch.
The design round from requirements through storage choices to tradeoffs.
The language patterns interview problems need, including the standard library shortcuts.
Built for candidates whose blocker is expressing a correct answer clearly in English.
A full programme when you want the whole loop covered rather than one round at a time.
Every HireCade course in one place.
A blank canvas for drawing system design answers while you practise talking through them.
Run Python in the browser with no installation, for testing an algorithm quickly.
If you are practising interviews but not getting them, the resume is usually the real blocker.
The whole path in order: learn, practise, build a resume, apply, interview, and negotiate.
Compensation ranges by role and company, for the conversation after the interviews.
How titles map to levels, which tells you how hard a loop will push.
