Remote PCB Design Engineer for AI Evaluation Tasks

Mercor; San Francisco (CA); Remote; USD 120,000 - 180,000; Full time

Job summary

Mercor is partnering with a leading AI lab to measure how frontier AI models perform real engineering design work. We're hiring experienced CAD, PCB, and 3D design engineers to build realistic, difficult design tasks that expose where AI still falls short.

You will design a realistic engineering problem (a prompt, input files, and a grader that checks the result), run it against frontier models, and refine it until it is fair, realistic, and hard. Realism always comes before difficulty.

Qualifications

• 3+ years hands-on design experience in at least one track (industry, research lab, or serious freelance work).
• Portfolio of your own design work (portfolio link required to apply).
• Proficient in Python and comfortable with the command line and Git.
• Ability to judge whether a design is actually correct, not just how it looks.

Responsibilities

• Turn real design work into a prompt, input files (STEP, KiCad projects, .blend files, meshes), and an automated grader.
• Build and iterate on tasks with an AI coding agent and version them in Git.
• Run each task against frontier models, read transcripts, and close leaks until the task is hard for the right reasons.
• Pass automated QA and peer review, and document the rationale behind any synthetic data you create.
• Attend onboarding and calibration sessions.

Skills

Python programming

Git version control

3D design

Hands-on CAD design

Portfolio ownership

Tools

SolidWorks

AutoCAD

FreeCAD

Siemens NX

CATIA

build123d

CadQuery

OpenCASCADE

KiCad

OpenEMS

Blender Python API

PyVista

VTK

kicad-cli

pcbnew Python API

Job description

Mercor is partnering with a leading AI lab to measure how frontier AI models perform real engineering design work. We're hiring experienced CAD, PCB, and 3D design engineers to build realistic, difficult design tasks that expose where AI still falls short.

You will design a realistic engineering problem (a prompt, input files, and a grader that checks the result), run it against frontier models, and refine it until it is fair, realistic, and hard. Realism always comes before difficulty.