Rehearse the interview. Then read the room live.
An AI interviewer asks, you answer out loud, it grades you. On the day, it captions what the interviewer says and drafts a reply in your own words, from your own projects and numbers.
Runs in Chrome. Uses your own AI key, so your profile and transcripts never pass through anyone else's server.
Two ways to use it
The same app, before the interview and during it.
Before: a mock interview that talks back
Paste the job description into your profile and press Interview Me. The AI plays the interviewer for that role. It asks one question, you answer out loud, you press Enter, and it tells you what landed, what was missing, and one thing to fix. Then it asks the next one, or digs into your last answer the way a real interviewer would.
- Behavioral, situational, and "walk me through how you'd fix it" questions
- Grades against the frameworks below, not against vibes
- Export the session and review it the night before
Interviewer
Tell me about a time a launch went wrong and what you did.
You, out loud, then Enter
We launched a referral program expecting fifteen percent of new merchants from referrals, and after three months it was three. I had assumed merchants talk to each other. Interviews showed they see each other as competitors, so I killed it and moved the budget…
Grade. Clear failure story, good ownership. Missing: the concrete cost of running it three months, and what you check now before building an incentive. Say the rule you adopted.
Next. How did you decide three months was long enough to call it?
Next. How did you decide three months was long enough to call it?
During: captions, then an answer in your voice
Play the call through your speakers and press Start Listening. The interviewer's words collect in a dashed box while they talk. When they finish, you press Enter: the question is sent, the mic mutes so your own answer isn't captured, and a reply streams in that you can read in your own words.
- Answers use your real projects and your exact numbers, never invented ones
- Names and jargon the transcript mangles get repaired from your profile
- Paste a screenshot when they show you something in writing
Heard
so walk me through how you would diagnose that if say one region's conversion is way below the others
Suggested answer
Before I touch anything I'd want two things: did it start on a date, or has it always been low? And is the metric measured the same way in that region as in the others?
Then I'd lay out the likely causes before testing any of them…
Then I'd lay out the likely causes before testing any of them…
Answers with a shape interviewers recognise
Every answer follows a structure chosen for the question type. You see the structure. The interviewer just hears a good answer.
STAR, spoken
Situation, task, action, result, as sentences you can say, not sections you announce.
- 1Situation. Where you were and what was broken, with a number.
- 2Task. What you were responsible for.
- 3Action. What you actually did, and one thing you found out along the way.
- 4Result. The before and after, in your own numbers.
Sample answer from the fictional profileMerchant onboarding. When I took it over, onboarding took nine days on average and 40% of merchants who started never finished. Sales blamed the form, compliance blamed missing documents, and nobody had the numbers.
I instrumented every step and found 60% of the drop-off was at document upload, mostly on mobile. So we split the flow: merchants could start selling with a limited account while documents were checked in the background.
Onboarding went from nine days to two, drop-off from 40% to 18%, and that came to about 3,000 extra active merchants a quarter.
Seven beats for a disagreement
The interviewer is quietly asking whether you would describe a disagreement with them fairly.
- 1Stakes. What it would cost to leave it unresolved.
- 2Sides. Both positions, both reasonable, each right about its own metric.
- 3Signal. The data you went and got before pushing.
- 4Reframe. The question you changed. This is the beat that matters.
- 5Resolve. The decision and the mechanism that made it acceptable.
- 6Repair. Who owned the call, and why the relationship survived.
- 7Rule. The durable fix so it never comes back.
Sample answer from the fictional profileCompliance wanted every new merchant fully verified before a single transaction. Sales wanted instant activation. While it stayed stuck, we were losing about a third of sign-ups in the waiting period.
Both were right about their own number. Compliance is measured on fraud losses, sales on activations. Before pushing, I pulled the data: 92% of fraud came from merchants above a certain monthly volume.
So I changed the question from "verify first or activate first?" to "how much can an unverified merchant safely do?" We put a transaction cap on unverified accounts instead of a full block. Compliance owned the cap number, so it stayed their decision. Fraud losses stayed flat, activations went up 25%, and we wrote the cap into policy so the argument never came back.
Diagnose before you fix
For "X is broken, walk me through it." The one answer type where going long is the point.
- 1Scope. Two sharp questions back, then your assumptions.
- 2Hypothesize. Six to eight causes out loud before testing any.
- 3Clarify. Why each cause would produce this exact signature.
- 4Segment. The data cut that confirms or kills each one.
- 5Contain. Stop the bleeding before the root cause is known.
- 6Fix. Tied to the confirmed cause.
- 7Prevent. The systemic change, so it can't recur anywhere.
Sample answer from the fictional profile, opening onlyTwo questions first. Did the drop start on a date, or has that region always been low? And is conversion measured the same way there, same funnel, same definition?
If you don't have those, I'll assume it's a step change in the last few weeks. Then the causes I'd want on the table before testing any of them: a release that only hit that region, a payment method that stopped working, a pricing or tax change, a tracking break, a traffic-source shift, or a local competitor move.
While we work through those, I'd contain it: pause any experiment running in that region, and put the region on a daily check so we see the moment it recovers…
Short lines you can glance at
For "imagine…" and "what would you do if…". Lines, not paragraphs, every one speakable as written.
- 1What matters most here, said naturally.
- 2One assumption, only if the question depends on things you can't know.
- 3Two to four steps, each one sentence.
- 4A closing link to something you really did, if one fits.
Sample answer from the fictional profileThe main thing for me is to find out whether it's real before anyone starts fixing.
I'd want to know if it started on one day or drifted across the week, and whether it's every country or one.
First I'd check that tracking didn't break, because a step change on one day is very often a release.
Then I'd cut the numbers by platform, country, and traffic source to see where the drop lives.
After that I'd look at what shipped that week and roll back the most likely change while we confirm.
This is basically what we did when onboarding drop-off jumped after a form change.
Nothing to trust us with
It's one file that runs in your browser. There is no account, no server, and no database on our side.
- Your profile
- Saved in your browser's local storage on your computer. We never receive it.
- Your API key
- Also local. Each answer goes straight from your browser to the provider you chose. You pay them directly, usually a few cents an answer.
- Transcripts
- Live only for the session. Refresh the page and they are gone, unless you export them yourself.
- Payment
- Handled by Lemon Squeezy at checkout. We never see card details.
Works with the key of your choice:
Anthropic ClaudeOpenAIGoogle GeminiDeepSeekOpenRouter
One price, yours to keep
$29one time
A download you own. No subscription, no account, updates included.
- Mock interviews with grading, for any role you paste in
- Live captions with a suggested answer in your voice
- Your own profile, projects, numbers, and voice
- Five AI providers, bring your own key
- Feedback chips that rewrite an answer and remember the rule
- Screenshot paste, session export, keyboard-only flow
Buy for $29
Try the demo first
Requires Google Chrome or Microsoft Edge and an API key from one of the providers above. AI usage is billed by the provider, typically under a dollar per interview.
Digital download, no refunds. The demo is the full app with a sample profile, so you can test your microphone, your key, and the answer quality before paying.
Questions people ask
Do I need to be technical to set it up?
No. You paste your background into a form, pick an AI provider, and paste a key. Getting the key means creating an account on the provider's site and adding a few dollars of credit. The app walks you through it step by step, with a "Test connection" button that tells you if something is wrong.
How does it hear the interviewer?
Through your microphone. Play the call through your speakers instead of headphones, and the mic picks up the interviewer's voice. Transcription runs inside Chrome and is free. It is not perfect, especially with names and jargon, which is why the app repairs those from the terms in your profile.
Will the answers sound like a robot?
They sound like your profile. You choose a voice: plain English for non-native speakers, natural conversational, or formal. The AI uses only the projects and numbers you wrote, and you can correct any answer with one click and turn that correction into a permanent rule.
What does it cost to run?
The app is a one-time purchase. Each answer sends your profile plus the question to your AI provider, which costs roughly one to five cents depending on the model. A full hour of practice is usually under a dollar. You set a spending limit on the provider's site so it can never surprise you.
Is using it during a real interview allowed?
That depends on the employer, and some tell candidates not to use assistance. Our view: use the practice mode as much as you like, and treat the live mode as captions plus a memory aid for your own stories, the way a non-native speaker or a hard-of-hearing candidate would. Read the interviewer's instructions and make your own call.
Which browsers work?
Google Chrome and Microsoft Edge, on Mac or Windows. Safari and Firefox lack the live transcription the app depends on.
Can I get a refund?
No, which is why the demo is the full app. Test your microphone, your key, and the answers on the sample profile before you buy. If a download is broken or you cannot open the file, email us and we fix it.