5-7 years of experience in evaluation, measurement, or data quality for machine learning systems.
Strong background in machine learning and data science, particularly with Python.
Proficient in statistics, including sampling, precision, recall, and confidence intervals.
Experience in designing evaluation protocols and ground-truth pipelines from scratch.
Ability to critically assess and challenge data presented by research teams.
Responsibilities
Own and define benchmark methodology for evaluation sets and metrics.
Write and align customer quality specifications and audit rubrics with rejection criteria.
Conduct human audits and collaborate on sampling and throughput with operations.
Develop automated checks and grading systems to minimize manual review.
Monitor and report on data quality, making decisions on data readiness for shipping.
Benefits
Collaborative work environment across multiple teams.
Opportunity to influence quality standards in machine learning.
Engagement with cutting-edge technology in robotics and data science.
Encouragement for applicants who may not meet every qualification.
Full Job Description
The role
When a robotics team trains on Mecka's data, they need to trust what they're getting. You'll define what that trust rests on: the benchmarks, audits and quality standards behind every delivery. You'll dig into the failures that averages hide, question results that look too good and make the call on whether data is ready to ship.
You'll work across Computer Vision, annotation, operations and delivery. We're looking for someone with a machine learning and data science background who enjoys getting to the bottom of a difficult measurement problem, and can turn what they find into a quality bar the whole team can use. What you'll do
Own the benchmark methodology: which evaluation sets exist, how they are sampled, which metrics apply and how versions are frozen and compared.
Write customer quality specifications and audit rubrics, and keep both aligned with what customers actually reject.
Run human audits and work with annotation and operations on sampling, agreement, throughput and cost.
Build automated checks and graders that reduce manual review costs and are hard to game.
Track where recorded hours are lost, report quality results regularly and make the ship or no-ship call on data.
What you need
Experience in evaluation, measurement or data quality for machine learning systems, or in data operations at scale.
A machine learning and data science background, with strong Python and working knowledge of statistics: sampling, precision and recall, agreement and confidence intervals.
Experience designing evaluation protocols and ground-truth pipelines from scratch.
The judgment to challenge a research team's numbers and support that challenge with evidence.
An eye for how a metric can be gamed, and the habit of closing those gaps.
Nice to have
Benchmarks and metrics: 3D pose, trajectory and reconstruction metrics such as MPJPE and ATE.
Audits and data quality: Sampling design, inter-annotator agreement, loss accounting and managing annotators or vendors.
Automated checks and graders: Verifiers, learned scoring, active learning and human-in-the-loop evaluation.
Data curation and scoring: Quality scoring, filtering and composition of large datasets.
If you don't check every box above but believe you can do the job, we encourage you to apply - we're looking for capability and trajectory, not a perfect checklist match.