Install the DevMesh extension.
Run any assessment in your IDE.
$ code --install-extension devmesh.devmesh-challengesFor engineering teams hiring in 20260+ assessments done
Hire engineers who ship with AI.
Candidates work on a real problem in their own VS Code with a DevMesh AI extension. We capture every prompt and event, then a deterministic engine ranks how well they steered AI and the calibre of what shipped.
- Runs in their own IDE
- 22 scorers · 6 judges · 7 anomalies
- Evidence-cited verdicts
The Assessment Engine
They build with AI.
We grade the receipts.
DevMesh runs in their IDE, not ours. We capture every prompt, edit, and decision. Then a deterministic, evidence-cited engine scores how well the candidate steered AI and the calibre of what they shipped.
Install the DevMesh extension.
01Their IDE
They build where they actually work.
DevMesh ships as a custom VS Code extension with AI built in. Candidates solve a real problem in their own editor. No toy sandbox, no babysat playground.
02Capture
Every prompt. Every event. Timestamped.
Each AI prompt, file edit, paste, retry, and terminal run is hashed and recorded. The whole session becomes a deterministic, replayable record.
03Rank
A two-tier engine grades the work.
Twenty-two deterministic scorers, six LLM judges, seven anomaly detectors. Each signal cites the exact prompt or file behind it. This is the moat.
Why teams switch
We score the work. You defend the call.
Two-tier scoring across twenty-two deterministic scorers, six LLM judges,
and seven anomaly detectors. Hash-pinned, reproducible, defensible.
The Unified Report
Final Score, Four Categories. 22 Hiring Signals
Each candidate's run collapses into a single comparable score and four category breakdowns: Judgment, Output, Process, and Efficiency. Behind every band is a cited piece of evidence, not a vibe.
Final score
Breakdown
- Judgment· 6 dimsGood86
- Output· 7 dimsGood94
- Process· 5 dimsAverage71
- Efficiency· 4 dimsAverage63
They use their own IDE
Ship a custom VS Code extension with AI built in. Real environment, real keystrokes, no babysat playground.
Calibrated across roles
Weighted dimensions normalized against role and seniority, so your hiring committee can compare apples to apples.
Auditable verdicts
A defensible "hire / no-hire" call with receipts attached. Every signal cites the exact prompt or file behind it.
See it on a real role0+ assessments done
Hand us a JD.
We'll show you the verdict.
We'll assemble a DevMesh assessment around the role, run a benchmark candidate, and walk you through the evaluation in twenty minutes. No commitment, no slide deck.
- 01Send us a JD
A real engineering hire you are actively working on, with the messy bits intact.
- 02We assemble the assessment
A custom challenge wired to your stack. The DevMesh extension ships with AI baked in.
- 03A benchmark candidate runs it
They solve the problem in their own VS Code. We capture every prompt and event live.
- 04You debrief with the verdict
A defensible call your hiring committee can sign off in five minutes. Every score cited.