AI that checks, not chats

Interview what the résumé claims.

Everybody’s résumé says React.Winkle’s AI is the part that checks.

Free to start. No card.

The pattern

Hiring is still a guess with paperwork.

Every line is an assertion with nothing attached to it, and the only source is the person being assessed.

Everyone calls this a process.

The idea

One evidence record. Two views.

They build one record. You read the same record from the other side. Nothing is re-entered, nothing taken on trust.

Ask the copilot or search anything…K
6
VO
Valentine OffiahEmployer
Talent

Amara Okafor

3/6 verifiedAvailable

Senior backend engineer (NestJS/Postgres) · she/her · Lagos, NG

Remote · hybrid $60—90 /hr Notice: 4 weeks Open to relocation amara@example.dev CV

At a glance

Strong signals

Senior backend engineer (NestJS/Postgres) · 3/6 skills verified · verification passed · scored 82 against Senior Backend Engineer · available now.

Verified

3/6

Checks

pass

Fit

82

Interview

strong

Assessment

88%

Status

Available

About

Eight years on payments and ledger systems. Most recently led the migration of a monolith to event-driven services without a scheduled outage.

Verification

pass94% confidence

Fit to role

82

Strong fit

3 verified · 1 gap

NestJSPostgreSQLTypeScriptKafkaKubernetes

Links

Portfolio GitHub

Skills · 3 verified

NestJS· expertTypeScript· expertPostgreSQL· strongKafka· competentgRPC· competentTerraform· familiar

Looking for

A backend team that ships to production weekly and treats on-call as a design problem.

Open to roles

Senior Backend Engineer Platform Engineer

Preferred locations

Lagos, NG Remote (EMEA)

Experience

  • Senior Backend Engineer

    Paystack · 2021–present

    Ledger and settlement services. Cut reconciliation time from hours to minutes.

  • Backend Engineer

    Andela · 2018–2021

    Built and ran payment integrations across four markets.

The employer console, reading one candidate’s evidence record. Illustrative — a rendering of the interface with invented data, not anyone’s record.

They claim. The system verifies.

A skill flips to verified when the AI finds it demonstrated — with the source and date attached. Verification is not the candidate’s to grant.

You read what is already proven.

Discovery ranks on verified skill. The interview spends its hour on the rest.

Most hiring tools stop at the offer letter.

That is the point where the work starts. Winkle carries one record from the first search to the hours you approve at the end of the month. Eight steps, one system.

  • 01

    Step 1 of 8Sourcing

    Discover.

    Opted-in people, ranked on verified skill.
  • 02

    Step 2 of 8Evidence

    Verify.

    AI reads the transcript and corroborates the claims.
  • 03

    Step 3 of 8Screening

    Interview the gap.

    The AI interviews only what is still unproven and grades it.
  • 04

    Step 4 of 8Decision

    Score.

    Every rating cites a timestamped moment, not a memory.
  • 05

    Step 5 of 8Commitment

    Offer.

    Fit scored on evidence. Gaps named before you commit.
  • Most hiring tools stop here
  • 06

    Step 6 of 8Contract

    Sign.

    Redlines, approvals, e-signature, executed PDF.
  • 07

    Step 7 of 8Delivery

    Engage.

    The contract becomes a live engagement the day it executes.
  • 08

    Step 8 of 8Money

    Approve hours.

    Weekly timesheets. Approve, dispute, reopen.

Hire on evidence. Not on a good hour.

Four things carry most of the weight. Open one.

AI verification

Skills that had to be earned.

Verification run

append-only

  • TypeScriptclaimedverified
    • 41 commits · 3 repos
    • explained the retry fix on camera
  • PostgreSQLclaimedverified
    • schema review · 2 PRs
    • named the index that fixed it
  • Kafkaclaimedunverifiable
    • no corroborating source found

Consent on file. A failing check degrades to unverifiable — it never invents a verdict.

AI interview

An interview that doesn’t waste the hour.

Interview plan

60 min

Already proven — not asked

TypeScriptPostgreSQLREST designTesting

Unproven — the hour goes here

  • Kafka2 questions
  • Incident response2 questions
  • Leading a team1 question

43 of 60 minutes on the 3 things nobody has proven yet.

Fit scoring

Fit, with the arithmetic shown.

Fit against the role

must-haves 70%

  • TypeScriptverified · 1.0
  • PostgreSQLverified · 1.0
  • Reactclaimed · 0.6
  • Kafkamissing · 0.0

0

Fit

Verified weighs 1.0, a claim 0.6, missing nothing at all. You can argue with this number — that is the point.

Contracts

Contracts that close here.

Contract lifecycle

executed

Signed · both partiesExecuted
  • Template
  • 2 redlines
  • Approvals 2/2
  • Signed
  • Executed PDF

Here is one, graded.

Every score points at the moment it came from. Click a line and the recording jumps there — so a verdict is something you can check, not something you take.

Ask the copilot or search anything…K
6
VO
Valentine OffiahEmployer
Interviews/ Senior Backend Engineer/Pipeline
Interview review

Tunde Adeyemi

completed · tunde@example.dev

Recommended

AI score

84

Integrity

96

Answered

6/6

Time spent

14:22

Flags

1

Winkle asks

Question 4 of 6

You said the ledger migration ran without a scheduled outage. Walk me through how writes were kept consistent during the cutover.

We dual-wrote to both ledgers behind a feature flag for two weeks, reconciled nightly, and only cut reads over once the drift report came back empty three days running…

Next question
Screen
Face
View
Side by sideScreenFace
Q1Q2Q3Q4Q5Q6
9:47 / 14:22
AI audio Next flag
0.75×1×1.25×1.5×2×

Transcript & timeline

Plays in sync — click any line to jump there

  1. 2:41Q2 answeredcorrect · 92

    How did you keep writes consistent during the ledger cutover?

    We dual-wrote behind a flag for two weeks and reconciled nightly before moving reads.

  2. 8:56Looked away from screen · 4sminor
  3. 9:14Q4 answeredcorrect · 88

    What broke first when the event bus fell behind?

    Settlement lag. We shipped a back-pressure guard the same week and alerted on queue age, not size.

Per-question analysis

Graded against each rubric · click to jump to the moment

  • 1. How did you keep writes consistent during the ledger cutover?

    correct · 92
    scenarioadvanced · Distributed systemsRubric

    We dual-wrote to both ledgers behind a feature flag for two weeks, reconciled nightly, and only cut reads over once the drift report came back empty three days running.

    Names the mechanism, the verification and the exit condition. Dual-write plus a drift check is the answer the rubric asks for, and the three-day gate is specific enough to be real.

    148s Jump to moment
  • 2. Where does that approach break down at ten times the volume?

    partial · 61
    probeadvanced · Distributed systemsRubric

    Reconciliation gets slower, so you would probably run it more often and shard the job by account range.

    Right instinct, no arithmetic. Doesn't reach the write-amplification cost of dual-writing, and sharding by account range is asserted rather than argued. Worth a follow-up question live.

    71s Jump to moment

By skill

Average score across questions tagged with each skill

  • Distributed systems
    88
  • Postgres
    82
  • Debugging
    79
  • Trade-off reasoning
    61

CV vs. interview

Where the CV holds up — and where it doesn't

Aligned

  • Migrated a monolith without a scheduled outage described the dual-write and the drift gate unprompted (aligned)
  • Owned settlement services named the failure his team shipped the fix for (aligned)

Worth a look

  • Led the platform team spoke as a contributor throughout; no scope or headcount given (unsupported)

AI recommendation

Recommended

AI assists — you decide.

84

Your decision

Advance Hold Reject

Ledger answers held up against the CV. Good fit for the settlement rewrite alongside Amara.

Notes save as you type

Integrity

Flags only — never an auto-reject

96
Looked away 4s, at 8:56

Summary

Answers stayed concrete under follow-up on distributed writes and on the failure his team actually shipped a fix for. The one soft spot is cost reasoning at scale — worth ten minutes live rather than another round.

Strengths

  • Reasons from what he measured, not from what he read
  • Names the verification step without being asked for it
  • Answers stay concrete at every level of follow-up

Gaps

  • Trade-offs at scale are asserted rather than costed
  • Leadership scope on the CV went unevidenced

One graded interview — the recording, the moments, and the score that came out of them. Illustrative — a rendering of the interface with invented data, not anyone’s record.

Everything else that is already built.

8 areas, 42 capabilities. Open any one of them.

  • Sourcing & talent

    Find people on proof, not keywords.

    5 capabilities
  • Roles & applications

    Post it, publish it, collect it — in one record.

    6 capabilities
  • Screening & interviews

    The hour goes on what nobody has proven yet.

    6 capabilities
  • Evidence & decisions

    Every call traceable to something checked.

    5 capabilities
  • Contracts & signing

    Drafted, argued, approved and executed here.

    6 capabilities
  • Work & time

    From signature to the hours you approve.

    6 capabilities
  • Team & permissions

    Who sees what, down to compensation.

    4 capabilities
  • Platform

    The plumbing, already plumbed.

    4 capabilities

Work & time, for instance.

A contract executes and an engagement opens. Hours are tracked against it, on projects, with tags and budgets — then approved into the invoice.

Ask the copilot or search anything…K
6
VO
Valentine OffiahEmployer
Time tracking

Refresh-token rotation on the auth endpoints

Winkle Mobile AppDevelopment
01:24:06
0%

Weekly goal

0m / 40h

40h 00m to go this week

9 days

Logged today — keep it alive

2/7

days tracked

This week

Week total · 10:05:26

Today

04:42:26

Refresh-token rotation on the auth endpoints

Winkle Mobile AppDevelopment
Running01:24:06

Ledger reconciliation service

Payments PlatformBackend
9:12 AM – 11:38 AM02:26:00
3

Review PRs #482, #486, #491

Winkle Mobile AppCode review
00:52:20

Yesterday

05:23:00

Timesheet approvals — API and UI

Winkle Mobile AppDevelopment
10:04 AM – 1:47 PM03:43:00

Approvals flow — design review

Design SystemDesign
2:30 PM – 3:25 PM00:55:00

Sprint planning

Internal· non-billableMeetings
4:00 PM – 4:45 PM00:45:00

The same console after the hire — time tracked against the contract you signed in it. Illustrative — a rendering of the interface with invented data, not anyone’s record.

It records. It never decides. That stays yours.

  • Opt in, or stay invisible.

    Not opted in, not in the search. No back door.

  • Consent, and a record of it.

    Every run written append-only. You can always show your working.

  • Verification is not self-service.

    Only demonstrated work flips a skill to verified. The source is stored with it.

  • Flags. Never verdicts.

    The record notes what happened and when, and you can jump to the moment and watch it. Nothing auto-rejects, nothing is scored as guilt, and no keystrokes are logged. A person decides, or nobody does.

The direction

Every verified skill is a data point about whether someone works out.

0

Fit

Today the dial is one candidate against one role. The evidence underneath it is the part that compounds.

Today that powers fit against a single role — must-haves weighted, gaps named, the arithmetic visible.

The direction is a Fit Score: portable, evidence-backed, and owned by the person it describes rather than by whoever happened to interview them.

We are building the record first. That part isn’t shipped, and we’ll keep saying so until it is.

The questions you were going to ask anyway.

Asked in the order people actually ask them, answered without the part where we avoid the answer.

Not hiring — looking?

Build the record instead of another PDF. Get skills verified once and carry the proof with you.

Hire on evidence. Not on a good hour.

Start hiring on evidence

Free to start. No card.