HUMAN DATA · CLEO · EXPERT INTERVIEW

Find the right expert. Prove who they are.

Cleo reads the brief, shortlists real specialists with reasons, and runs the interview. The verdict, the transcript, and the verified identity travel together — so the proof of who did the work is there from the first conversation.

Talk to Human Data See pricing See Cleo run

CLEO · INTERVIEW SESSION

BRIEF

Read once

Must-haves, constraints, and tone, in your words.

SHORTLIST

Ranked with reasons

Why each specialist surfaced. What they bring. What they don't.

VERDICT

With the proof attached

Structured interview. Same rubric. Transcript and verified identity, signed.

HOW IT WORKS

Bring the brief. Keep the proof.

One brief in. A verified specialist out, with the verdict and the proof of who they are attached.

STEP 01

WHAT WE READ

Bring the brief

Cleo takes the brief, the must-haves, and the real constraints, then turns that into a search and interview plan.

→

STEP 02

WHAT WE RUN

Get a verified specialist

Profiles are ranked by fit, each with the reason it surfaced. Cleo runs the structured interview and scores the rubric live — the same rubric, every candidate.

→

STEP 03

WHAT WE SIGN

Keep the proof

Hire, no-hire, or human review — with the transcript, the rubric, and the verified identity attached. The first link in a chain of consent that survives an audit.

PROVENANCE · IT STARTS AT THE INTERVIEW

Who the expert is travels with them.

Signed means something here: an identity-verified, audit-surviving chain of consent attached to the person who did the work. It begins at the first conversation.

AT INTAKE

Identity is verified and signed

Before a specialist reaches the brief, their identity, credentials, and consent are verified and signed. Not a self-reported profile — a checked, dated record of who they are and what they agreed to.

Identity · credentials · signed consent → The first link in the chain

THE SWITCH

Onboarding experts like consumers stopped being acceptable

When a competitor lost four terabytes — including who its workers were and how it ran RLHF — buyers stopped trusting unverified supply. The answer is a neutral source you can defend under audit, with identity checked at the door.

Verified at intake → A second source you can defend

ON THE RECORD

Every verdict carries its evidence

The transcript, the rubric, the reason a candidate surfaced, and the human override if one was made — all stay on the record. No one retypes a summary. The record is the handoff.

Transcript · rubric · routing reason · override → One record, signed

AT EXPORT

When the regulators ask, the answer starts here

78% of organizations cannot validate where their training data came from and 77% cannot trace it. When the data this expert creates is exported, the answer to who made it and under what rights begins at this interview — provenance starts at intake.

Who created it · who reviewed it · under what rights → Audit-ready provenance

EU AI ACT

Provenance that survives August 2026

High-risk training-data provenance enforces in August 2026, and provenance includes the person who created each datapoint. That chain has to start somewhere it can be defended. It starts at the qualification step Cleo owns.

Verified person · signed consent · attached proof → Survives the audit

RANKING · EVERY RANK HAS A REASON

The ranking is configurable. No black box.

Ranking weights are stored in config. Teams review and sign off before the weights change. Every shortlist comes with the weights that produced it — and the same weights, scored the same way, give the legal and talent team an adverse-impact trail they can review.

FACTOR 01

40%

Semantic similarity

How close the candidate's experience reads to the brief — not keywords, but meaning across role and domain.

FACTOR 02

25%

Experience score

Years of work, depth of programs, and the texture of the projects on file. Real work, weighted.

FACTOR 03

20%

Skill coverage

Coverage against the explicit skills the brief calls out, including adjacent ones the role tends to need.

FACTOR 04

10%

Availability

Whether they can start when the team needs them. Confirmed, not assumed.

FACTOR 05

Referral signal

Internal references, prior calibration, and second-degree signal from the bench.

EXAMPLE · EXPERT DATA ROLES

Candidate A.P.

Clinical NLP specialist

Oncology

Privacy review

NLP

AVAILABILITY GATED

Candidate M.N.

Genomics data reviewer

Genomics

Rubrics

2 WEEKS

Candidate L.A.

Safety & bias analyst

RLHF

Bias

Ethics

AVAILABILITY GATED

Anonymized example. This is the work: clinical NLP, genomics, safety and bias — the expert data frontier labs and regulated programs staff. Teams configure weights per role before search runs.

ROUTING · THREE PATHS · ONE RECORD

Three routes out of every session. The evidence travels with each one.

Cleo does not pick a route in the dark. The rubric score, the availability check, and the review signals stay attached to the record the route was decided on.

PATH 01 · TO INTERVIEW

Route to a structured round

Signal is strong enough for a structured round. Cleo schedules the interview session and attaches the rubric, the transcript, and the reason for the route.

SIGNALS

· Strong rubric fit

· Availability confirmed

· No review flag

PATH 02 · TO ASSIGNMENT

Direct to calibrated work

Candidate is ready for calibrated work directly. Cleo hands them into Workforce with their verified identity and the cleared rubrics attached.

SIGNALS

· Prior calibration on file

· Skill taxonomy match

· Workforce opening flagged

PATH 03 · TO HUMAN REVIEW

Escalate with the evidence

Edge case, review signal, or mixed rubric. Cleo pauses, keeps the evidence together, and brings in the recruiter on call.

SIGNALS

· Review flag triggered

· Rubric contradictions

· Novel role profile

RUBRIC · REVIEW SCORECARD

The rubric moves as the session moves. So does the call.

Dimension-weighted rubric updates during the interview, not after cleanup work. The hiring team sees how a candidate is doing while the session is still running.

Notes stay on the row that earned them. The score doesn't arrive without the reason. Same rubric, every candidate — the call is reviewable, not a black box.

REVIEW SCORING · DURING THE SESSION

SCORECARD · EXAMPLE

WEIGHTED · IN PROGRESS

Technical depth

4.2 / 5

WEIGHT · 35%

Strong on distributed systems; gaps in GPU optimization

Problem solving

3.8 / 5

WEIGHT · 25%

Clear reasoning; needed hints on edge cases

Communication

4.5 / 5

WEIGHT · 20%

Structured answers; good use of examples

Domain knowledge

3.5 / 5

WEIGHT · 20%

Solid NLP foundations; limited clinical exposure

PROGRAM NOTE · ON THE RECORD

“The first round used to eat half the team. Now Cleo runs it, the rubric stays the same across every candidate, and the recommendation — with the verified identity attached — is ready before the candidate logs off.”
Head of talent operations · representative talent program

WHAT COMES OUT

What the brief turns into.

Every session leaves the proof the team needs to defend the call — and the next interview can compare against.

Shortlists

A ranked list with the reason each specialist surfaced and what they don't bring.

↳ ARTIFACT

Rubric reads

Scoring across the dimensions the brief calls for. The same rubric, every candidate, reviewable after the fact.

↳ ARTIFACT

Interview transcripts

The full session, with the moments that moved the score marked and timestamped.

↳ ARTIFACT

Signed identity

Identity, credentials, and consent verified at intake — the proof of who did the work, attached to the verdict.

↳ ARTIFACT

Hire / no-hire verdicts

One read, with the rubric and the transcript attached. Routes ambiguous cases to a human.

↳ ARTIFACT

ADMIN GOVERNANCE

The interview an audit can defend.

Before the candidate record hands off to Workforce or your ATS, five controls answer the questions legal and talent ask about an AI-conducted interview. The state of each one stays attached to the session.

DISCLOSURE

The candidate is told it's AI

Before the session starts, the candidate sees the persona, the interview type, and the rubric. Transparency is the policy, not a workaround.

↳ ACTIVE

FAIRNESS

Same rubric, reviewable

Every candidate runs the same rubric with weights signed off in advance. The scoring is consistent and the adverse-impact trail is there for legal and talent to review.

↳ ACTIVE

OVERRIDE

Human override is logged

The recruiter on call can override any route Cleo recommends. The override stays on the record with the reason and the rubric snapshot at the time.

↳ LOGGED

SYNC

Workforce + ATS handoff

The verdict, transcript, rubric, and verified identity move as one packet into Workforce or your ATS — with a delivery record on each handoff.

↳ PASSING

ABUSE REVIEW

Bad sessions route to a human

Sessions where the abuse classifier pauses the conversation are surfaced for human review with the full transcript and the reason.

↳ REVIEW

QUESTIONS · BEFORE YOU RUN IT

What teams ask before they run Cleo.

Six questions, answered plainly. The same six come up on every program — recruiting, talent ops, and the program lead all want the same thing.

01 · ON THE RUBRIC

Whose rubric runs the interview?

Yours. Cleo runs the rubric your team has already been using. The rubric is versioned, the changes go through approval, and every session gets stamped with the version it was scored on.

02 · ON THE TRANSCRIPT

Where does the transcript live after the session?

On the candidate record, with the rubric, the routing decision, and the reason. The transcript is portable — it leaves the workspace as part of the handoff packet, not as a separate export.

03 · ON THE ROUTE

Can a human override Cleo's route?

Yes. The recruiter on call can override any route Cleo recommends. The override stays on the record with the reason and the rubric snapshot at the time it was made.

04 · ON THE PERSONA

Is the candidate told they are talking to AI?

Yes. The candidate sees the persona, the interview type, and the rubric structure before the session starts. The transparency is the policy, not a workaround.

05 · ON THE BACKUP

What happens when voice fails?

The session continues in text. The rubric does not reset. The transcript merges the voice portion and the text portion under one record, with the failover timestamp marked.

06 · ON THE HANDOFF

Where does a successful candidate go next?

Into Workforce, with the cleared rubrics, the transcript, the route, and their verified identity attached. The recruiter does not retype a summary — the packet moves as one record.

WHAT CHANGES

What changes when Cleo runs the first round.

Frontier labs have run out of internet-scale data. What they need now is expert human data — coding, math, law, medicine, finance — quoted by field and seniority. Finding and qualifying real specialists is the hard part, and it breaks when search, outreach, and the first-round interview live in four tools. By the time a decision is needed, the evidence has already split, and nobody can prove who the expert was.

Cleo collapses that into one conversation. The brief is read once. The shortlist comes with reasons. The interview runs with the same rubric every candidate. The recommendation is ready from the session record, not after cleanup. And the proof — every transcript, every rubric, every reason, and the verified identity — stays with the specialist as they move into reviewed work.

WHO THIS IS FOR

Built for staffing expert data work.

Frontier labs and regulated programs staffing real specialists — not crowd labor. The fields where a wrong answer costs, and where who created the data has to hold up under audit.

Coding & agents

Math & finance

Clinical & medical

● EXPERT DATA

Law

Safety & bias

RELATED MODULES