NAX Ai Interviewer is a voice-first screening interviewer that replaces your entire first round. You send a single-use link; the candidate opens it whenever they are ready; an AI interviewer greets them, asks role-specific questions out loud, listens to the spoken answer, follows up where the answer is thin, and closes the session with a scored verdict — all without a human in the room.
Everything a recruiter normally does in a 20-minute phone screen is automated: verifying the basics, probing real experience, running a short live task, watching for dishonest behaviour and writing the summary. What lands on your desk is a shortlist with recordings, transcripts, per-answer scores and an integrity log attached.
No forms and no multiple choice. The candidate speaks in natural language and the AI responds in the same language, turn by turn, the way a real interviewer would.
No calendars, no reschedules, no no-shows. A candidate can finish at midnight and your team reads the verdict at 9am.
Every candidate gets the same questions, the same think time and the same rubric — so the comparison between two applicants is actually fair.
One continuous take, timestamped integrity events and a full transcript. Nothing is reconstructed after the fact.
Every feature exists to make hiring decisions faster and more accurate — for your team and for the candidate.
Candidates speak naturally. The AI listens, understands context and follows up like a human interviewer — no forms, no typing.
English, Bahasa Malaysia, Chinese or a natural rojak mix. The AI detects the language turn-by-turn and answers in the same one.
Data, Content, HR, E-Commerce or Live Host — questions and live tasks are tailored to the role being hired.
Brief network drop? The session pauses, buffers locally, and resumes on the exact question — nothing is lost.
Video and audio bitrate scale to the candidate's connection in real time so recording stays smooth start to finish.
Tab switches, extra monitors, copy-paste, multiple faces and long silence are logged. Five strikes ends the interview.
Screenshots, screen recording, right-click, view-source and DevTools shortcuts are blocked and blur the screen.
Continuous video from first question to last, compressed and uploaded resumably to a private, access-controlled bucket.
Answers scored on relevance, clarity and depth. A Pass / Review / Fail decision lands the moment the session ends.
Every interview gets a unique code. Candidates and HR can re-open status or results any time.
After each question candidates get up to 30 seconds to gather their thoughts — pressure-free by design.
Invite links are strictly one-time and expire on your schedule. No back-doors, no duplicate attempts.
A closer look at the machinery: how a question is chosen, how an answer is judged, and what evidence is kept behind every verdict.
Each role is broken into competencies — for a live host that might be energy on camera, product recall, objection handling and shift reliability. Every question maps to one competency and carries its own model answer and scoring anchors. Because the anchors are written before anyone applies, two candidates answering the same question are measured against the same yardstick rather than against each other's charisma.
When an answer is short, generic or misses the competency the question was written for, the interviewer asks one targeted follow-up before moving on. That single probe is what separates a rehearsed line from real experience, and it is capped so a nervous candidate is never interrogated.
The candidate is never asked to commit to a language up front. If they open in Bahasa Malaysia, slip into English mid-sentence and return to Bahasa, the interviewer follows them. Scoring is applied to the substance of the answer, not to accent or fluency, so a strong operations candidate is not penalised for imperfect English.
Tab switches, window blur, a second display appearing, clipboard activity, more than one face in frame and unusual silence are each written to a timestamped log as they happen. Five events end the session automatically. Nothing is inferred afterwards — a reviewer can open the log, jump to that second of the recording and judge the moment themselves.
There is no stop, retake or edit. Video and audio are captured from the first question to the last, buffered locally if the network dips and uploaded resumably to a private bucket with retention you set. Transcript timestamps link straight into the video, so a reviewer can verify a claim in seconds instead of watching twenty minutes.
Every session ends with Pass, Review or Fail plus a per-question breakdown across relevance, clarity and depth, a written summary and the reasoning behind the call. Your team keeps the decision; NAX simply removes the guesswork and the first-round workload from it.
No booking slots and no queue — open a role and every applicant can sit the interview at the same time. One candidate or five hundred in parallel, the session runs the same.
Sub-second turn-taking, so the conversation feels like a call and not a form.
Local buffering and resumable upload survive a dropped connection mid-answer.
Recordings live in a private, access-controlled bucket with retention set by you.
Identical questions, think time and rubric for every candidate in the same role.
No black box. Every number on the verdict card comes from something you can point at in the transcript.
Did the answer address the question that was asked, or a nearby question the candidate would rather answer? This is where most weak answers separate themselves, and it is the easiest part of the score for a reviewer to verify by reading two lines of transcript.
Could a colleague follow the answer the first time? We look at structure and specificity, not polish — a plain answer with real detail beats a fluent one that says nothing.
Did they show the reasoning behind the answer, or recite a definition? Follow-up questions exist mainly to test this, which is why the AI pushes once on anything that sounds rehearsed.
Every score is a recommendation with the evidence attached. Reviewers change verdicts, and when they do, we use that to tune the rubric for the next batch instead of arguing with it.
Candidates should never think about the technology. That only happens when the unglamorous parts — buffering, retries, storage, access control — are handled properly.
Answers are buffered locally as they are spoken and uploaded in resumable chunks. A dropped 4G signal mid-sentence pauses the upload, not the interview, and the recording is still complete when it lands.
The whole interview is a web page. It works on mid-range Android, on an old iPhone, on a shared laptop in a cyber cafe. No installs means no candidate lost to a download that failed.
Recordings, transcripts and resumes sit in a private bucket behind access-controlled links. Nothing is public, nothing is indexed, and retention is a number you choose rather than one we assume.
The session is recorded as a single continuous take with the integrity log stitched to the same timeline, so anyone reviewing later can see exactly when something happened rather than taking a summary on trust.
The interview service, the scoring service and storage are health-checked continuously and reported on the public status page. If something degrades, you read it there before a candidate emails you about it.
Question banks, prompts and the interface are pre-translated, not machine-translated on the fly. That keeps the wording consistent for every candidate in a language and keeps latency low for all of them.
Tell us the roles you hire for and we'll configure a pilot for your team.