Litmus

Engineering teams waste time writing interview problems, running screening calls, and grading take-homes, yet still don't get a real read on how a candidate builds before meeting them. Traditional take-homes are easy to game and don't show process: how someone uses AI tools, iterates on code, or explains their decisions. Teams need a high-signal, work-trial-like read on technical ability without burning engineering hours, and without breaking their existing ATS workflow.

Litmus generates async coding assessments scoped to a team's own context, producing realistic tasks plus an automated grader and an agent baseline run. Candidates complete end-to-end feature builds in their native setup with their own AI tools (Claude, GPT, etc.) available, while Litmus tracks file iteration, commit history, and every AI prompt used. Submissions run against a deterministic harness in a live sandbox for correctness, and configurable LLM graders evaluate qualitative dimensions like architecture and process, with judgments tagged to specific files and lines. Candidates also record a short walkthrough explaining their decisions, and results compile into a shareable report. Litmus syncs with ATS platforms like Ashby, Greenhouse, and Lever, or can run the full pipeline end to end, and can also host and grade a team's existing take-home materials.