I want to describe a moment you already know.
You've prepared for a week. You've read the job spec until you could recite it. You walk in, you sit down, you're fine — and then someone says five words: "So, tell me about yourself."
And there is nothing there.
Not panic exactly. Something quieter and worse. The answer you rehearsed forty times in the car has simply left, and in its place is a kind of white noise, and the longer the silence runs the louder it gets. You hear yourself say "basically" and "fundamentally" and you know, even as you're saying them, that they aren't words. They're filler while you wait for your brain to come back.
It comes back about ninety seconds after you leave the building.
I've watched people I know are excellent do this. Not underqualified people. Engineers with twelve years of production experience who freeze on a question about their own résumé.
Here's the thing I eventually understood: they didn't fail because they didn't know the answer. They failed because it was the first time they'd ever said it out loud.
Knowing something and being able to say it under pressure are different skills. We train the first one obsessively — certifications, courses, documentation. We train the second one essentially never, and then we're surprised when it collapses the one time it's tested.
I'm a network and security engineer. For years people have messaged me the week before an interview — can you run me through some questions, can you look at my CV. I always said yes. And I could always only help one person, once, for an hour, and only if I happened to be free that evening.
That bothered me for a long time before I did anything about it.
The hardest problem was teaching it to say nothing
So I built the thing I kept wishing I could hand people instead. An AI interviewer that has read your CV and the job you're going for, that speaks your language, that asks the follow-up question rather than the next one on a script.
I expected the difficulty to be the intelligence. It wasn't. The models are good enough. The hardest engineering problem, by a distance, was teaching it to wait.
Here's why that's harder than it sounds. Every voice pipeline ever built is optimised to detect the end of your speech as fast as physically possible. That's the whole latency game — the industry competes on shaving milliseconds off "the human has stopped talking, respond now." Endpointing, it's called, and everyone wants it faster.
But a nervous candidate goes quiet constantly. Mid-sentence. Mid-word. Not because they're finished — because they're thinking, or because they've just heard themselves say something clumsy and they're deciding whether to start again.
Every tool I tested interrupted them at precisely that moment. Which is the single most destructive thing you can do to someone who is already losing their nerve. It doesn't just break the sentence. It teaches them that hesitating is punished, so they start filling every silence with noise, which is exactly the habit you're supposed to be curing.
Getting an interviewer to sit calmly through a twenty-second silence and then gently offer a nudge, rather than cutting in, meant fighting that behaviour at the client, at the flow orchestrator and in the prompt layer at the same time. Three layers all engineered to do the opposite of what I needed.
It's the least visible feature in the product and it took the longest. It's also the first thing every single person notices.
Then I let it interview me
Building something for other people is comfortable. At some point I ran a full session on myself — senior network role, my own field, my own CV.
It gave me 70 out of 100.
Then it told me to go and get my CCIE and my CISSP.
That stung for about a day. It was also completely correct, which is the annoying thing about honest feedback. I'd been telling other people to prepare properly while quietly assuming I didn't need to.
I kept the score in the product exactly as it is because of that. Not a pass mark. Not encouragement. A number, the specific answers where you lost the room, and what to fix before the real thing.
The second thing I learned was about CVs
The other half of what I built analyses CVs, and it taught me something I genuinely did not expect.
Most CVs aren't weak. They're written for the wrong country.
A CV that works perfectly in one market gets filtered out in another before a human ever opens it. A photo, a date of birth, a marital status — completely normal in some places, an instant automated rejection in others. Duties listed where outcomes should be: "responsible for network operations" instead of "cut outage time 40% across 14 sites." Two pages of skills and not a single number.
None of the people I've seen hit this were underqualified. Every one of them was invisible.
And nobody tells you. The rejection email certainly doesn't — you just get silence, and you conclude the problem is you. It usually isn't. It's a formatting mismatch you were never told about, and it takes about ten minutes to fix once someone points at it.
If you're job hunting right now
Three things, and you don't need my product for any of them:
Say your answers out loud. Not in your head. Out loud, to a wall if necessary. The gap between the answer in your head and the answer in the room is the entire problem, and you can only find it by hearing yourself.
Put numbers in your CV. Replace every "responsible for" with something measurable. If you can't measure it, you probably haven't described the outcome yet.
Find out what your target country's CVs actually look like. Not a generic template — the real convention where you're applying. This one change does more than a week of applications.
What I built, honestly
ShortlistMe is live. A patient AI interviewer, in your own language, that's read your CV first and waits while you think. An honest scored report after every session. And a CV studio that scores and rewrites for the format your target country expects.
It's free to start — two CV analyses and one five-minute session, no card. Paid plans start at $19, and the first fifty members get 40% off for life with the code FOUNDING50.
What it won't do is promise you a job. Nobody can promise you that, so I don't. What it does is make sure the first time you say your answer out loud isn't in the room that matters.
If you're preparing for something right now, write to me at info@shortlistme.ai with your target role. I'll look at your CV myself.
— Muhammad Waqar, Melbourne
Illustrative images generated with AI.