ML Engineer, Calibration

We publish our accuracy. Dated predictions, graded outcomes, misses listed by name. That record is the company's main claim, and this role keeps it honest.

What you’ll own

• The Scorecard: backtests, forward registers, and the kill-gates that retire a signal.
• Pre-registration. Thresholds get committed before the test runs, and they hold.
• The grading pipeline: outcome labels, cohort definitions, and the choices that make a number mean something.
• Retiring signals that don't clear the bar, in public, with our name on it.
• The methodology write-ups a skeptical customer will read closely.

What you’ll bring

• You've built evaluation for models in production, not just trained them.
• Comfort with calibration, base rates, and why AUC alone doesn't settle an argument.
• Willingness to publish a miss. This work is adversarial by design.
• Clear writing. Half the job is explaining a result to someone who wants to disbelieve it.

Logistics

Location

Washington, DC or remote (U.S.)

Authorization

U.S. work authorization required

Engagement

Full time