Intervues · Product
An interview engine is software that conducts a structured interview itself — deciding what to ask next, adapting to a candidate's answers, scoring responses, and producing evidence a hiring team can check — rather than just scheduling, recording, or storing an interview someone else conducted.
Ask three recruiters what "interview engine" means and you'll get three different answers, because the term is new enough that nobody owns it. Here's the specific claim: an interview engine is the layer that actually runs a structured interview — not the layer that stores candidates, not the layer that finds them, and not the layer that gives them a written test.
Running an interview means several things happening together, in real time, while a person is talking: capturing what they say, deciding whether to let them keep talking or to interject, deciding what to ask next based on everything they've said so far — not just the next question on a fixed script — noticing when an answer contradicts something said ten minutes earlier, and, at the end, turning a conversation into a score that's backed by specific, checkable moments in the transcript rather than a single number nobody can explain.
None of that is what an ATS does, none of it is what a video interview tool does, and none of it is what a resume-screening AI does. Those tools are valuable, but they sit around the interview. An interview engine is the thing that conducts it.
An applicant tracking system is a system of record: it stores candidates, tracks which stage of a pipeline they're in, and lets recruiters leave notes and move people between columns on a board. It does not decide what question to ask a candidate, and it does not listen to what they say.
Most "AI" features bolted onto ATS platforms today are resume parsing and keyword matching — useful for sorting a pile of applications, but they never actually talk to a candidate. An interview engine can plug into an ATS the same way a human interviewer's notes do: as an input to the pipeline, not a replacement for it. The ATS still tracks who's in the pipeline and at what stage; the interview engine is what runs one of the stages.
A one-way video interview tool asks a candidate to record themselves answering a fixed list of questions, alone, with no one — human or otherwise — actually listening. A recruiter watches the recording later, usually on fast-forward.
That's fundamentally a recording tool, not an interviewing tool. It can't follow up on an interesting answer, can't ask a candidate to go deeper on something vague, can't notice when someone contradicts an earlier answer, and can't tell the difference between a candidate who pauses for three seconds to gather their thoughts and one who has no answer at all — because nothing is listening in real time. An interview engine's whole job is to be present in the conversation: deciding, live, whether to wait, prompt, or move on, the way an attentive human interviewer would.
"AI recruiter" usually describes something that operates before the interview: sourcing candidates, writing outreach messages, parsing resumes, ranking applicants against a job description. That's a sourcing and matching problem — finding the right people and getting them into a pipeline.
An interview engine picks up after that: once a candidate is in front of it, the sourcing question is already answered, and the job becomes evaluating whether this specific person can actually do the job — through conversation, not a resume-match score. A hiring stack can use both: an AI recruiter to build the pipeline, an interview engine to evaluate it.
Coding tests, psychometric assessments and skills quizzes are static: every candidate gets the same fixed set of items, answers independently, and gets scored against a fixed rubric. That's a legitimate way to test some things — syntax knowledge, working memory under time pressure, personality inventories — but it can't adapt to what a specific candidate says, and it can't test the skill that fails silently on a resume: whether someone can explain what they know out loud, under the mild pressure of a real conversation.
An interview engine is adaptive by design — the next question depends on the last answer, not a fixed sequence — which is what makes it suited to testing communication and reasoning under conversational pressure, rather than testing a checklist of facts.
Underneath the term "interview engine" is a specific stack of problems that all have to be solved together, because a conversation doesn't pause for engineering. Here's what each layer actually does:
Capture
Turning a candidate's microphone (and sometimes camera) into a clean stream the rest of the system can work with — including detecting when they've started and stopped talking, which is harder than it sounds over a patchy connection.
Turn-taking
Deciding, continuously, whether to keep listening, jump in, or wait. The hard case is silence: is a three-second pause someone thinking, or someone finished? Get this wrong and the engine either talks over people or leaves them hanging.
ASR (speech-to-text)
Converting speech to text accurately enough that the reasoning layer isn't working from garbage — which is markedly harder across accents than most speech systems are tuned for by default.
Reasoning (the "interview brain")
Deciding what to ask next based on everything said so far, not a fixed script — while holding memory across the conversation so it can notice a later answer contradicting an earlier one.
Scoring
Turning a transcript into a structured judgment against a rubric, applied the same way across every candidate for a role — not a vibe, and not a single unexplained number.
Evidence
Attaching every score to the specific, citable moment in the transcript that produced it, so a hiring manager — or a candidate disputing a score — can check the claim against what was actually said.
Write-back
Getting the result — score, evidence, transcript — into wherever the hiring team or candidate actually looks: an ATS, a dashboard, a profile. An evaluation nobody downstream can see isn't useful.
A structural comparison, not a scorecard — each category of tool is built for a different job, and most hiring stacks end up using more than one.
| Primary job | Talks live? | Adapts in real time? | Produces evidence? | |
|---|---|---|---|---|
| Interview engine | Conducts the interview | Yes | Yes | Yes |
| ATS | Tracks the pipeline | No | No | No |
| Video interview tool | Records fixed answers | No | No | Rarely |
| AI recruiter | Finds & ranks candidates | No | No | No |
| Assessment platform | Tests fixed items | No | No | Partial |
Two specific engineering problems make this harder than it looks, and they're worth naming because they're the ones that quietly break interview engines built for only one market.
The first is accent robustness. Speech-to-text systems built and tuned mostly on native US or UK English commonly transcribe Indian-accented English far less accurately — word error rates several times higher than on the accents they were tuned for. For an interview engine, a bad transcript doesn't just look messy: it's the input the reasoning layer works from, so a garbled transcript produces a wrong or confused follow-up question, which the candidate reasonably experiences as the system not listening. Solving this isn't optional for a product built for a global, accent-diverse candidate pool — it's the difference between an engine that works for everyone it claims to serve and one that quietly works for a subset of them.
The second is patience. A scripted IVR system fills every silence instantly, which trains people to feel rushed. A good human interviewer does the opposite: they let a short pause sit, offer a little more time if it stretches, and only step in to rephrase if someone seems genuinely stuck — because thinking time is not the same as not knowing the answer, and confusing the two is one of the most common ways interviews penalize people who process out loud more slowly, including many second-language speakers. Building that judgment into a turn-taking layer — treating a three-second pause differently from a thirty-second one — is a deliberate design decision, not a default any speech system arrives with.
It's worth being specific about where this particular engine actually stands, because "interview engine" is easy to say and much harder to ship honestly. The reasoning, turn-taking, ASR and scoring layers described above have been built and tested — including against real spoken sessions — but the engine has not yet been released as a live interview that anyone can simply take today. Continuing to test it before opening it up matters more than moving fast on this specific piece, because a candidate's first real interview shouldn't be the first time the system is finding its own edge cases.
That's a narrower claim than "we have an AI interviewer," and it's the accurate one right now.
Questions
No. An applicant tracking system is a system of record that tracks candidates through a pipeline; an interview engine is the layer that actually conducts a structured interview. Most hiring stacks use both — an ATS to track the pipeline, an interview engine to run one stage of it.
It's built to handle the structured, repeatable part of screening — the same first-round questions asked consistently at volume — so human interviewers can spend their time on the calls that need judgment a machine shouldn't make alone, not to remove people from hiring decisions.
Because the reasoning layer only ever sees what the speech-to-text layer transcribes. Speech recognition tuned mostly on native US or UK accents transcribes other accents far less accurately, and a bad transcript produces a bad follow-up question — which a candidate reasonably experiences as the system not listening.
Deciding, in real time, whether a pause means someone is thinking or finished. Interrupt too early and the engine talks over people; wait too long and candidates feel abandoned mid-answer. Getting this right means deliberately modeling silence in tiers rather than reacting to every pause the same way.
A citable link between a score and the specific moment in the transcript that produced it — so a hiring manager, or a candidate disputing a result, can check the claim against what was actually said, rather than trusting an unexplained number.
The engine — reasoning, turn-taking, ASR and scoring — has been built and tested, including against real spoken sessions, but it has not yet been released as a live interview anyone can take. We're continuing to test before opening it up.
Yes — it's designed to plug in as one stage of a pipeline, feeding its output (score, evidence, transcript) into whatever system a team already tracks candidates in, the same way notes from a human interview round would.
Intervues is the ecosystem this engine plugs into — practice, evidence and a profile candidates keep.
· 3 free credits · pay per interview · nothing recurring