It scores the interview. Then it hides the score from you.
Orin runs the structured first round against your competency framework, records and transcribes it, and evaluates afterwards — then holds that evaluation until your panel has submitted their own. Show a panel a number first and their judgment bends toward it. So they don’t see one.
- Your panel decides first
- Asks for a person, it stops
- Recordings auto-delete
- 01Check the file is read and structured before the interviewInternal
- 02Run the structured first round against your competency frameworkExternal
- 03Evaluate afterwards, and hold the evaluation sealedInternal
- 04Turn what it found into probing questions for the panelInternal
- 05Write the evaluation to the employee record after a hireInternal
An AI score shown too early replaces the interview it was meant to support.
Anchoring isn’t a weakness some interviewers have. It’s what happens to people. Put a number in front of a panel before they deliberate and their assessment moves toward it — including the panel members who would tell you, sincerely, that it didn’t.
What you end up with is an AI decision with three human signatures on it. You’ve paid for the panel’s time, you’ve kept the appearance of human judgment, and you’ve quietly stopped having any.
The fix is not a better model. It’s sequence. The panel goes first.
Five tasks, each labelled with what it can reach.
Turn any of them off individually.
Check the file first
Confirms the CV has been read and structured before the interview, so questions are grounded in what the person actually wrote.
Never: interviews on an unread file.
Run the first round
Conducts a structured interview against the skills in your competency framework, with follow-up questions generated from the answers given.
Asks first: before recording anyone — separate consent.
Evaluate, and hold it
Produces a level per competency with evidence pinned to the recording — and keeps the whole evaluation sealed until your panel submits.
Never: releases early, for anyone, including admins [CONFIRM].
Turn the held score into questions
Converts what it found into suggested probing questions for the panel — the gaps worth testing, without revealing the conclusion.
Never: shows the score inside the question set.
Write it to the employee record
If the person is hired, the interview evaluation is stored with them so it can be compared with real performance at ninety days.
Never: writes anything for a candidate who wasn’t hired.
Nothing here can close a candidate’s application. That class of action exists and always belongs to a person.
Your panel submits first. Then the AI’s opinion unlocks.
Every competitor sells the opposite — instant scores, waiting for you when you open the file. That is exactly the arrangement that makes a panel’s judgment worthless, and we’d rather demo badly than ship it.
The interview happens.
Recorded, transcribed, structured against your framework.
The panel interviews and submits.
Independently, without seeing anything from the AI.
The evaluation unlocks.
Only when the last panel member has submitted.
Two opinions, side by side.
Yours first, then a second one with its evidence attached.
What this costs us, and why we did it anyway.
Instant scores demo better. A buyer clicks once and sees a number, and that feels like the product working. Holding it back means the first thing a prospect sees is a lock. We accepted that, because a panel that has already been anchored is a panel you’re paying for and no longer using — and because a hiring decision that can’t be shown to have been made by people is a decision that’s hard to defend to anyone who asks.
The hold applies to everyone with access, including administrators. There is no override. [CONFIRM this is true before publishing it.]
Not a number. A better set of questions.
Holding the score back doesn’t mean the panel walks in unprepared. The agent converts what it found into suggested probing questions — the two or three areas where the first-round answers were thin, phrased as things to ask rather than conclusions to confirm.
“Ask about a time they led a design end to end” is useful. “System design: level 3” would have decided the conversation before it started. The first version helps a panel interview better; the second replaces them.
The questions never contain the score, the level, or a hint of the direction. That’s checked as part of how they’re generated. [CONFIRM]
- Areas to probe, not conclusions
- No score, level or direction leaked
- Generated from the first-round answers
One sentence from the candidate ends it.
If a candidate says they’d rather speak to a person, the interview stops there. Not a form to fill in, not a request that gets reviewed, not a note added to a file. It ends, and the role’s recruiter is notified to take it over.
Nobody has to justify the request or explain why. Being assessed by a machine is something a person is entitled to decline, and a right that requires an argument isn’t one.
It’s also the honest test of everything else on this page. A product that means what it says about human decisions will let a candidate opt out of the machine at any moment, at no cost to them.
- Stops immediately, mid-interview
- No reason required
- Their application is unaffected
No problem. We’ve let the hiring team know, and someone will contact you to arrange a conversation. This doesn’t affect your application.
Nothing from this session is scored. [CONFIRM whether a partial session is discarded or retained — and say which, plainly.]
Told plainly, and it costs them nothing.A level per competency, pinned to the second it came from.
Questions come from the skills in your competency framework, and results come back as a level per skill against the level that role requires — the same five-level scale as your assessments and your performance reviews.
Every judgment carries an evidence pin: a timestamp and the sentence that produced it. Click it and the recording jumps there. A number you can’t interrogate is a number you can’t defend to the candidate who asks.
Competencies the interview didn’t reach are shown as not covered, rather than scored as absent.
- Levels, not a percentage
- An evidence pin on every judgment
- Gaps in coverage shown, not filled in
You’re recording a person’s face and voice.
Under Vietnam’s Personal Data Protection Law, biometric data is sensitive personal data. It needs its own explicit consent — not a clause inside a general application form — and a retention limit you can state. [CONFIRM WITH COUNSEL]
Separate, explicit consent.
Asked on its own screen before the interview begins, naming what is recorded, why, who can see it, and for how long. Declining routes the candidate to a human interview — it doesn’t remove them from your process. [CONFIRM WITH COUNSEL, and confirm the decline path exists]
A stated retention limit.
Recordings delete automatically after [CONFIRM] days, configurable. The candidate is told the number before they start, and it appears next to the evaluation rather than in a settings page.
A named audience.
Only roles you’ve granted — the recruiter and the panel by default. Every view is logged.
Deletion on request, from either side.
Early deletion records a reason so the log stays meaningful. [CONFIRM WITH COUNSEL]
What it doesn’t assess.
It doesn’t score emotion, personality, confidence, enthusiasm, body language, tone of voice, or anything described as cultural fit. Those are proxies for age, health and background, and there is no version of them we’d want to defend. It assesses the competencies in your framework and nothing else.
This describes how the product works and is not legal advice. Our consent flows are reviewed by Vietnamese counsel. [CONFIRM WITH COUNSEL]
The evaluation doesn’t stop mattering on the hire date.
When someone is hired, the interview evaluation goes onto their employee record. At ninety days their actual performance is assessed on the same five-level scale, and the two are compared.
Which produces the answer nobody in this market currently has — not whether AI interviewing works in general, but whether it works in your company, for your roles.
If the relationship turns out to be weak, that’s what the report will show. We’d rather hand you a number that undermines our own scoring than have you discover it in year three.
- Evaluation follows the hire
- Compared at 90 days on the same scale
- Reported to you, either way
- Requires Tanca HRM
Four sources, four switches, one log.
Read-only. Nothing writes back.
Each one is a separate switch. An agent doing three things you want beats one doing four where one makes you uneasy.
Interviews use your monthly allowance.
Shared across all six agents.
Draws on your AI allowance
- An interview a candidate ended early doesn’t count [CONFIRM]
- Recordings and evaluations stay available at zero remaining
- Credits are self-serve
Unlimited and free
- Reviewing evaluations
- Watching recordings
- Reading transcripts
- Evidence pins
- Panel briefings
- Sharing with your team
- Everything in Pipeline, Talent Pool, Assessments, Competency Framework, Pre-Onboarding and Reports
At zero remaining, new interviews pause. Nothing already recorded or evaluated is lost.
This is one of five.
Zola
Adapts and publishes roles across every channel, tracks cost per qualified applicant.
Learn more →Iris
Reads CVs against your competency framework and routes candidates to the right next step.
Learn more →Orin
Conducts a scored first-round interview against the role’s competency requirements.
You’re reading this oneRemy
Drafts and tracks offer letters, with salary band checks before anything goes out.
Learn more →Hana
Manages the checklist from signed offer to the end of probation.
Learn more →They share one allowance, one log and one queue.
Measured inside the product.
Food retail chain · 22 stores
of first-round calls removed per opening, per month [CONFIRM]
Logistics company · 3 warehouses in the south
from application to first human interview [CONFIRM]
Contact centre · recruitment hub
agreement between held AI evaluation and panel assessment [CONFIRM]
You’ll see the same figures in your own run log from the first role.
Questions people ask.
Why hide the score from us?
Because seeing it first changes what your panel concludes, reliably, including for people who are sure it doesn’t. The panel submits, then the evaluation unlocks as a second opinion.
Can an administrator unlock it early?
No. There’s no override. [CONFIRM]
What if only two of three panel members submit?
It stays sealed. You can remove a panel member from the requirement, which is recorded in the log. [CONFIRM]
What does the panel get before they submit?
Suggested probing questions generated from the first-round answers, with no score, level or direction in them.
What if a candidate wants a human?
The interview stops immediately and the recruiter takes over. No reason required, no effect on their application.
Does the AI decide anything?
No. It conducts, evidences and offers a second opinion. Stage-closing actions always belong to a person.
How long are recordings kept?
[CONFIRM] days by default, configurable, deleted automatically. Candidates are told the number before they start.
Do you assess tone, confidence or cultural fit?
No, and we won’t. It assesses the competencies in your framework.
How many interviews are included?
[CONFIRM] per month, shared across the six agents.
What happens to the evaluation if we hire them?
It goes onto their employee record and is compared with their performance at ninety days, on the same scale.
Let your panel go first.
A structured, recorded, evidenced first round — and an opinion that waits its turn.
Start free- Panel decides first
- Asks for a person, it stops
- Recordings auto-delete