I Built an Interview Coach Because Feedback Never Arrives in Time

"I do it on our daily sync every day"

I asked a backend candidate one question: could you give me an example of how you communicated with your team members to solve an issue.

He said, "I do it on our daily sync every day."

That was the entire answer, not the opening of one.

His hard skills were good. He was a nice guy. But I was hiring into a team with a lot of internal communication, and I'd just asked him to show me he could do the thing the role was mostly made of, and he'd handed me a calendar entry. I skipped him.

I don't think he was unqualified. The pattern I keep running into is people who can do the work and have never once had to describe it out loud, and that answer sounded like it.

The feedback exists. It usually doesn't reach you.

Candidates don't get to see what happens next. After that interview I wrote down why: it would be hard for him to communicate inside the team, and he should prepare for interviews. HR read it, checked it, and a version of it went out to him. Internally we also logged him as a no-go for now who might be worth another look later. He didn't get that part.

My version got polished on the way out. Polished means softened.

He got something back, which makes him the exception. A lot of companies just ignore that part, and you never hear anything at all.

It's rarely malice. They don't have the time, and they don't particularly want a relationship with the candidates they said no to, so the step gets dropped and nobody notices it's missing. And sometimes, from the inside, the honest version is simply too bad. Nobody wants to be the person who softens it enough to send, so it never gets sent.

There's a legal layer under it as well. Specific written reasons can be used to support a discrimination claim, so rejections tend to arrive as "we decided to move forward with other candidates" and nothing more. That vagueness is policy more often than it's laziness.

The result is the same either way. The reason exists in writing, in a system you can't see.

The truth stays on the company side

I've been on the receiving end of this too. You get the two-line email and you go back through the interview looking for the moment. It's impossible to conclude anything. You can guess, but the truth is on the company side and it stays there.

Something else took me longer to accept, from the other chair. Even the real reason isn't a fixed fact about you.

Soft skills are where hiring managers disagree with each other most. Every one of us has our own view of what the level should be, so it's always an argument. Two people can watch the same answer and score it differently, and neither of them is being unreasonable.

So no rejection email could have told you the answer, because there wasn't a single answer to give. There was a team, and a person comparing you against it.

One thing does hold steady underneath all that. The candidate answered a question that deserved a story with one sentence about his calendar, and every hiring manager in that room would have marked it down. That part isn't a matter of taste.

Why I built something instead of advising one candidate at a time

There wasn't a single moment where I decided. It accumulated over years of running interviews, watching the same failure from the same kind of person, and thinking every time that it would be great if people had an instrument to prepare with.

There was a selfish reason as well, and I'd rather say it than pretend the motive was pure. A prepared candidate makes my interviews better. They get to the information I need faster and more clearly, without the frustration on both sides, and they're ready when I follow up. The hour is more useful to me too.

The first version took a few days to build. It did roughly what it does now, just worse.

It was never a tool for engineers. It was built for everyone from the start, and the first users who weren't me proved the point: mechanics, a few teachers, medics. None of them were preparing for a system design round.

That's worth sitting with if you write software for a living. The failure that sinks a backend candidate is the same one that sinks a nurse. You know the work, you've just never had to describe it to a stranger under time pressure, and describing it turns out to be a separate skill from doing it.

What it actually does

You paste the job posting URL, or just type the title if you don't have a link. It parses the posting. You can upload a resume, paste one in, or skip it entirely.

Then you're in a session, out loud, in the browser. Three questions, twenty minutes. Practice runs push-to-talk, so you decide when you're answering and you're not fighting a voice detector while you think. Full mock interviews use automatic turn detection instead, which means the interviewer can cut in, the way one actually does.

At the end you get a score out of five, a readiness rating, and per-skill feedback. The feedback is built on quotes: it pulls two to five things you actually said and puts a rewritten version next to each one, so you get your own sentence back improved instead of an instruction to be more specific.

Readiness is tracked across sessions rather than judged from one transcript, because a single bad session is noise and a flat line across four of them is not.

The two things I got wrong

The first was that it interrogated people. It would take one topic and dig, five or more follow-ups deep, long past the point where a real interviewer would have moved on. That isn't rigor, it's a machine that doesn't know when it has enough. I capped follow-ups at three.

The second was worse. The scoring came out too strict and handed bad assessments to everybody, every user and every session.

A grader that fails everyone is measuring nothing. It's the rejection email that tells you nothing, with a number stapled to it. If the tool can't separate a good session from a weak one, it hasn't given you information, it's just made you feel bad with extra steps.

So I recalibrated the scale against the whole set of results rather than tuning it session by session. It's more accurate now, and I'd rather say that plainly than pretend the first version was fine.

What it can't tell you, and what to do first

When I give feedback as a hiring manager, it's specific to my team, my work, my context. That's what makes it worth having. The product's feedback is abstract by comparison, and I'm not going to pretend that gap is small. That's the limit of AI interview coaching, mine included. It doesn't know my team, so it can't tell you the role needed someone who'd be talking to other people all day.

What it can do is the part that's actually rehearsable. It puts you in the chair, makes you answer out loud, and shows you the sentence you produced. Most people have never heard themselves answer these questions, and the first time is a shock. Better to have that shock in your kitchen than in a real interview.

Before any tool, mine included, ask again. If you got the two-line email last week, reply and ask for feedback one more time. HR are busy people, not hostile ones, and a second ask sometimes gets a real answer out of a queue where the first one got buried.

If nothing comes back, stop trying to reconstruct the verdict from the silence. You can't, and I say that as someone who has both written those verdicts and sat waiting for one.

Go and answer the question again, out loud this time. If what comes out is one sentence about your daily sync, you've found the thing that was costing you the offer, and you found it somewhere it doesn't count against you.