Failure Analysis Researcher

The role The most useful thing we can tell a customer is where their model breaks and why. You’ll read agent rollouts at volume, build taxonomies of failure, and turn the patterns into the next round of tasks. What you’ll own Review of model trajectories across domains, with failures categorised rigorously.A failure taxonomy shared by dataset leads and researchers.Quantified shifts in failure modes between model versions and training runs.Task proposals that target persistent failures.The analyses that go to customers. What you bring An analytical background: research, data science, investigation or audit.Stamina for close reading and the discipline to code it consistently.Python or SQL for working with logs at scale.Strong writing. Helpful, not required Qualitative coding, error analysis or incident review.Domain depth in one of our program areas.