EN 中文
Discuss your first agent
All work JetDevs · Case study

We built the AI behind JetDevs' developer vetting.

JetDevs places remote engineers with fast-growing companies. Every candidate is verified, tested and matched before a client meets them, and we built the AI that does it at scale.

Client JetDevsTalent pool Asia PacificWho gets through Top 2%Profiles to the client In as little as 24 hoursInterview scoring ~15 minutes
Under the hood
  • Event trigger on application
  • Four-lane vetting board
  • Memory per candidate
  • Human spot-check lane
  • Identity and GitHub connectors
  • Learns from client hires
01 The challenge

Vetting that scales, with nothing fake getting through.

Before AI, grading one take-home test took a senior developer 2 to 4 days, and every video interview needed a person to watch it. JetDevs vets hundreds of partner agencies and the candidates behind them, and over 80% of companies worry about identity fraud in remote hiring. The vetting had to scale and stay strict.

02 What we built

Four checks on one Cadra board.

A new application fires an event trigger and lands on the board. From there it moves lane by lane: screen, skills test, interview scoring, match.

  • Starts on its ownAn event trigger starts vetting the moment an application arrives.
  • A lane for each checkScreening, the skills test, interview scoring and matching are lanes on one board, so every candidate's place is visible.
  • Candidate memoryPersistent memory keeps each candidate's checks, scores and evidence in one record.
  • A human spot-check laneSenior engineers review interview scores in their own lane before a profile moves on.
  • Cross-checks through connectorsIDs, references and GitHub, LinkedIn and portfolio footprints are checked through connected tools.
  • Learns from hiresSelf-learning: client hire feedback tunes how the next candidates are matched.
03 How it works

The board, lane by lane.

  1. 1

    Screen

    AI reads every resume.

    It flags inconsistent credentials and discrepancies. IDs, background checks, references and GitHub, LinkedIn and portfolio footprints are cross-checked.

  2. 2

    Source

    One standard for every source.

    Partner agencies are vetted too, and every candidate meets the same screening criteria, wherever they came from.

  3. 3

    Test

    Proctored tests, scored interviews.

    Live skills tests written by JetDevs engineers, screen-shared and webcam-proctored, with AI flagging suspicious behaviour. An LLM scores the video interview and senior engineers spot-check it. Then a 26-point character check.

  4. 4

    Match

    AI reads the client's brief.

    Matching puts only best-fit engineers in front of the client's final interview, straight into the client's Team Console.

04 Guardrails and control

People decide.

  • PeopleHumans check the AI. Senior engineers spot-check interview scores. Nobody is hired on a model's word.
  • EvidenceScored on evidence. Every score points to what the candidate said or did. Accent and origin are never scored.
  • IdentityReal people only. IDs, references and work samples are checked, and tests are proctored live.
  • ReplacementOne-for-one replacement. JetDevs stands behind every match and replaces the engineer if the client asks.
05 Outcome

What changed

~15 minto score a video interview with an LLM, then spot-check it
2–4 daysa senior developer used to spend grading one take-home test
Top 2%of developers across Asia Pacific get through
06 Stack and channels

What it runs on.

Client view
Team Console
Proctoring
Screen share and webcam
Character check
26 points
Sources
Agencies and direct, one standard
Discuss your first agent Next: ChainXP All work