Technical interviews have an uncomfortable problem.
You can know the answer and still give a bad interview.
You can understand databases, distributed systems, APIs, system design, or algorithms. You can have built real software and solved real production problems.
Then the interviewer asks: "Walk me through your approach."
And suddenly knowing something isn't enough.
You have to explain it.
Out loud.
Under pressure.
That is the part of interview preparation that most tools don't focus on.
The difference between knowing and explaining
Most technical interview preparation is built around knowledge acquisition. You solve coding problems. You study system design. You memorize common questions. You read interview guides. You build lists of things you need to revise.
All of that can be useful.
But an interview isn't a written examination. At some point, another person is sitting across from you and asking you to explain your thinking. That introduces a completely different skill. You have to:
- understand the question
- structure your thoughts
- explain your reasoning
- choose what information matters
- communicate trade-offs
- respond to follow-up questions
- notice when your answer is incomplete
- recover when you lose your train of thought
And you have to do all of that while someone is listening.
That's why interview practice should involve speaking, not just studying.
Why AI mock interviews are different from chatbots
AI has made it much easier to generate interview questions. But generating questions isn't the difficult part.
A real interview is a conversation. The interviewer listens to your answer and decides what happens next. If you mention a database migration, they might ask about it. If you make a questionable architectural assumption, they might challenge it. If you pause because you're thinking, they don't necessarily assume you've finished. If you contradict something you said earlier, they may come back to it.
That requires a system that can do more than display a question and wait for text. It needs a conversational loop: you speak, it listens, it decides, it speaks.
That is the fundamental idea behind a voice-based AI interviewer.
What makes a voice AI interview feel different?
A useful AI interview system needs to deal with the things that make human conversation messy.
1. It needs to listen. The system has to convert your speech into something it can understand. But speech isn't clean text. People pause. They restart sentences. They say "um." They change direction. They mix languages. They think while speaking. A useful voice interview system needs to account for that rather than treating every pause as failure.
2. It needs to know when you're finished. This sounds trivial until you try to build it. If the system responds too quickly, it interrupts you. If it waits too long, the conversation feels broken. That's why conversational turn-taking matters — a candidate should be able to pause and think without the interviewer immediately filling the silence. At Intervues, this is designed as a silence ladder: the system can hold, reassure, offer help, and only eventually redirect when appropriate. The goal is simple — thinking shouldn't look like stalling.
3. It needs to decide what to ask next. A question bank can give you another question. A conversational interviewer should be able to ask a better question because of what you just said. Imagine you're discussing a multi-tenant architecture and you mention that you partitioned a database by tenant_id. A generic interview system might move to its next predefined question. A conversational system can instead ask: "When you partitioned by tenant_id, what was the trickiest constraint on the write path?" That is a fundamentally different experience — the question comes from the conversation.
An AI interviewer should remember the interview
Memory is another important difference.
Suppose you say something at minute 14: "Writes were fully consistent." Then, seventeen minutes later, you explain that you dropped a constraint and reconciled writes later. A useful interviewer shouldn't have forgotten the first statement. It should be able to recognize the inconsistency and decide whether it is worth asking about.
That's the idea behind a contradiction ledger. The system can remember an earlier statement, connect it to something said later, and ask about the difference once — without turning the interview into a gotcha exercise.
It can also recall useful details from earlier in the conversation. You might mention a Postgres migration at minute 12. Later, when discussing scaling, that experience can become relevant, and the interviewer should be able to bring it back into the conversation. That's closer to how a good human interviewer behaves.
This is where Intervues comes in
I'm building Intervues, an AI system designed to conduct live spoken interviews. The core loop is: you speak, ASR, the interview brain, TTS, it speaks back.
But the interesting part is what governs that loop. Intervues is designed around:
- conversational turn-taking
- an Interview Director deciding what to ask next
- memory across the interview
- contradiction detection
- recall of earlier statements
- patient handling of silence
- barge-in, so speaking over the system stops it
- natural conversational backchannels
- Indian-English and code-mix calibration
The objective isn't to create another chatbot with a microphone attached. It's to create an interview conversation.
Why not just use an AI copilot?
Because interview preparation should prepare you for the interview. Not hide the difficulty of it.
There are already plenty of systems that can help someone generate an answer. That's not the problem I'm trying to solve. If an AI whispers the answer while you're being interviewed, you haven't necessarily become better at interviewing — you've just become better at using a copilot.
And that creates an uncomfortable question: if you need a copilot to pass the interview, will you need one to do the job?
Intervues takes the opposite approach. You have to answer. You have to explain. You have to think. You have to speak. The system's job is to create a realistic environment in which you can practise doing that.
Interview preparation should also be evidence-based
There's another problem with interview platforms: opaque scores. A candidate shouldn't simply be told System Design: 7.8/10. The obvious question is why — what did I actually demonstrate, where did that conclusion come from?
Intervues follows a broader principle: evidence over invention. A skill shouldn't exist merely because an AI model decided to label you with it. The system should be able to connect claims to evidence.
That philosophy extends beyond interview scoring. For example, a skill in a candidate's profile can be connected to the actual sentence in their résumé or repository that supports the claim. If the evidence doesn't exist, the system shouldn't manufacture it. The product deliberately refuses to show numbers it can't defend.
Who is Intervues for?
Intervues is initially focused on software and IT interviews, with India as the first market. That includes:
- students from tier-2 and tier-3 colleges
- early-career software engineers
- career switchers
- developers preparing for technical interviews
- engineers who know their subject but struggle to communicate under interview pressure
The focus is voice-led and Indian-English first. The idea is not to make candidates sound like someone else. It's to help them communicate what they actually know more effectively.
Where Intervues is today
There's an important distinction between what I'm building and what is already live.
The candidate product is live today — see pricing for credits. You can already see the candidate experience around profile and knowledge evidence, including graph building, quoted evidence, job-fit analysis, and the sample report.
The live spoken interview engine is a different story. The engine has been built and tested offline, but it has not been released as a live interview that anyone can simply take today. I'm one live microphone test from shipping that experience.
And I'm deliberately saying that. Build in public is only interesting if the public gets to see the actual build — including the unfinished parts.
The future of AI interview preparation isn't more questions
I don't think the future of interview preparation is another database containing thousands of interview questions. The harder problem is practising the conversation itself.
Can you explain your architecture clearly? Can you defend your decision? Can you respond when the interviewer changes direction? Can you recover when you lose your train of thought? Can you think for a few seconds without panicking because the other person is silent? Can you communicate what you actually know?
Those are skills worth practising. And they require something closer to an interview than a question bank.
That's what I'm building with Intervues — an AI system that conducts spoken interviews, not a chatbot with a timer.
If you're preparing for software or IT interviews and want to see what we're building, explore the candidate experience at Intervues. The product is being built in public, with one principle above everything else: Earned, not gamed.
Intervues is a members' club for honest, voice-led interview practice. Start practising → — or, if you'd rather see the work first, join the waitlist for the redacted sample report and the launch note.
Related reading
Questions
Why can a candidate know the material and still fail the interview?
Interviewing is a separate skill from knowing the answer. It requires structuring your thoughts, choosing what matters, communicating trade-offs, and recovering mid-sentence — all while someone is listening in real time. That translation skill only improves with spoken practice, not more reading.
How is a voice AI interviewer different from a chatbot with questions?
A chatbot displays a question and waits for text. A voice interviewer runs a live conversational loop — it listens, decides what to ask next based on what you just said, tracks contradictions across the session, and recalls earlier answers, the way a real interviewer does.
Does Intervues use a live AI copilot to help me during the interview?
No. Intervues is built for practice, not for cheating in a real interview. The system asks, listens, and pushes back — there's no whispering the answer to you. The goal is that you can answer without one.
Is the live spoken interview on Intervues available today?
The candidate product — profile, knowledge graph, job-fit analysis, and the sample report — is live now. The live spoken interview engine has been built and tested offline and is close to shipping publicly.