The recruiter
we always
needed.
An AI recruiter that shows its work: it argues each candidate from the evidence, cites every claim, names every gap, and puts the verdict first.
It slows down where hiring usually gets vague: gaps, dissent, and unsupported claims.
Advance to a final conversation. The operational evidence is real; the turnaround credit is shared; one gap must close before an offer.
Ran a 14-property portfolio through reopening; occupancy recovered ahead of the market curve.
Portfolio recovery is documented; her personal pricing authority is not. Closes with: a reference call to the group CFO.
SELF-ASSESSMENT — evidence coverage: Moderate · 2 of 4 claims independently corroborated · 1 gap open · interview minutes cited
CV matching answers the wrong question.
A keyword score can tell you a CV resembles the job description. It cannot tell you whether the claims inside it hold, what was carefully left out, or which decisions the candidate actually made when it mattered.
Interviews without evidence leave no record.
An unstructured conversation produces an impression — warm, articulate, confident. An impression cannot be audited, compared, or challenged later. Calibration needs a record.
We answer the question leaders actually ask: what does the evidence support, and what is still missing?
How the decision is made.
- I
Drop — or apply
A leader drops a CV and a job description — or candidates apply on a public job page. Consent is captured before any processing; every document is read and cited, never paraphrased from memory.
- II
The council
Three agents test each candidate from opposing sides: one builds the case for, one builds the case against, and one rules — with the dissent preserved verbatim.
Case forThree consecutive turnarounds, each verified against filings. CV §2 · filings
Case againstTwo of the three overlapped a sector-wide recovery. Only the third is a clean signal.
Case forThe third was a distressed asset — she set pricing herself. Interview 18:22
Case againstHer authority there is her own claim. No document shows it. Mark it unverified.
The rulingAdvance. The distressed-asset case stands; the authority claim is downgraded to unverified.
The rulingThe case against goes into the brief, verbatim.
Fictional exchange - III
Aria's interview
Aria runs a live voice interview against an evidence-driven agenda — no scripts, no generic competency questions.
Your CV says you led the turnaround. The board pack shows the margin recovering — it doesn't show the trade-offs. Which decision did you personally make that bent the curve?
Generated fromCV §4 — “led the turnaround”×Board pack p.7 — recovery shown, attribution absentEvery probe is written for this candidate, from their own evidence — and the holes in it. Senior interviews run 40 minutes.
Fictional example - IV
The brief
One page, verdict first. Every claim cited; every gap named with the channel that closes it. It lands in minutes, not weeks.
- V
The decision — and the loop
The leader decides. The decision is recorded against the brief, and a follow-up ninety days on feeds the closed-case registry.
+90-day follow-up scheduledDecision recordedHired — by the leader, on the recordThe outcome lands in the closed-case registry — the audit trail the next brief is graded against.
Fictional example
The Recruiter does not make the hiring decision. It prepares the evidence, preserves dissent, records the decision, and learns from the outcome.
One page. Written to be said out loud.
Plain language a decision-maker can use in the room — not analyst prose that needs translating.
Advance to a final conversation. The operational evidence is real; the turnaround credit is shared; one gap must close before an offer.
Ran a 14-property portfolio through reopening; occupancy recovered ahead of the market curve.
Held the flag partnership through an ownership dispute; the renewal was signed inside her tenure.
Portfolio recovery is documented; her personal pricing authority is not. Closes with: a reference call to the group CFO.
“The recovery rode a market-wide wave. Treat the occupancy curve as weak evidence of personal impact.”
SELF-ASSESSMENT — evidence coverage: Moderate · 2 of 4 claims independently corroborated · 1 gap open · interview minutes cited
The fastest way to judge the interviewer is to be interviewed.
Proof the brief is not just fluent.
It grades itself the way it grades candidates: every brief checked before it ships, every failure flagged, every closed case audited against what the leader actually decided.
9 automated checks every brief must pass
Format, citations, verdict placement, gap naming. A brief that fails any one of them does not ship quietly.
A tripwire, not a quarterly review
A non-compliant brief alerts the operator the moment it renders. Quality is enforced at the point of delivery, not discovered in an audit later.
Audited against real decisions
Every verdict is scored against the call the leader eventually made. The record: three closed cases, three direction matches — including one where the system overruled its own sceptic and the leader hired. Operating history, not a statistic; every new decision joins the trail.
It sharpens with every decision
The leader marks agree or disagree on every brief. A follow-up lands at ninety days. A weekly self-evaluation digest reads the record. And no prompt change ships without passing a replay of the closed cases.
Skeptical? Good. That is the posture the product is built on.
Candidates are people, not payloads.
One-time-code sign-in.
No passwords to leak. Every record is isolated to its tenant at the database layer — not by convention, by enforcement.
Expiring links. Consent first.
Candidate documents sit behind links that expire. Consent is captured before any processing begins.
Untrusted by default.
Everything a candidate supplies is treated as untrusted input to the AI. No personal data lands in the logs.
First deployed inside an executive hiring workflow at EMAAR; in live internal use as of July 2026. The operating requirement was simple: one page, verdict first, no invented numbers, every gap named.
Built for senior, high-consequence searches where evidence quality matters more than applicant volume.
A job description, the candidate's files, an approved consent flow, one operator, and a named decision owner. The system runs as an isolated tenant; prompt changes pass a replay gate before they ship.
How are candidates protected?
Consent is captured before anything is processed. Everything a candidate supplies is treated as untrusted input to the AI. Their documents sit behind expiring links, and no personal data lands in the logs.
Can the AI make the hiring decision?
No. It prepares the evidence, preserves the dissent, and records the decision a person makes. The verdict in a brief is a recommendation with sources — never the decision itself.
What happens when it's wrong?
The dissent was preserved, so a wrong call can be traced to what was argued at the time. The leader's disagreement is recorded against the brief. The closed-case registry keeps score, and the ninety-day follow-up exists to catch what fluency hides.