What the work involves

You author circuit design problems and evaluate model responses against them. In practice that means writing prompts drawn from work you have actually done — sizing a two-stage Miller-compensated op-amp for a given phase margin, debugging a switching regulator that oscillates under light load, resolving a setup-time violation in a pipelined datapath, choosing between an LDO and a buck for a noise-sensitive rail. Each item ships with a reference solution and a rubric explaining what separates a correct answer from a plausible-sounding one.

The grading half is harder than it sounds. Language models produce circuit answers that read fluently and cite the right equations while getting the pole locations, the bias point, or the units wrong. Your job is to catch that, say precisely where the reasoning broke, and rate the response against the rubric rather than against your own preferred design. Reviewers frequently disagree; expect adjudication threads where you defend a score with a load line, a small-signal derivation, or a simulation result.

What the screen looks for

  • Whether you have designed and taped out, bench-tested, or brought up real hardware — not just coursework or textbook familiarity
  • Depth in a named area: analog IC, RF, power electronics, high-speed digital, mixed-signal, or PCB signal integrity
  • Ability to explain a design decision with numbers, and to hold that explanation under follow-up questioning
  • Whether you can spot a subtly wrong answer that is superficially well-argued
  • Comfort with the tools you claim: Cadence Virtuoso/Spectre, LTspice, HSPICE, Altium, SystemVerilog, MATLAB

The screen is AI-conducted and probes claims rather than accepting them. Naming a project invites three follow-ups about the specification, the failure you hit, and how you fixed it.

Logistics

Fully remote and asynchronous. Contributors typically commit 10–20 hours a week against batch deadlines, chosen around their own schedule; some batches are shorter and more intense. Pay is hourly or per accepted task depending on the project, and observed rates in this band track specialisation and quality history rather than tenure. You need your own machine and, for some batches, access to a simulator — free tools such as LTspice are usually acceptable, and licensed tool access is a plus rather than a requirement.