Failure Analysis Researcher

Reinforced Learning; Town of Tonawanda (NY); Sur place; USD 70 000 - 120 000; Plein temps

Résumé du poste

Reinforced Learning in New York is seeking an analytical thinker to help identify where models break and why. You will read large-scale agent rollouts, build a taxonomy of failure, and translate findings into actionable tasks for the next development cycle.

The role requires strong writing, data analysis skills, and the ability to work with Python or SQL to process logs. You will collaborate with researchers and dataset leads to inform customers with precise, well-documented analyses.

Qualifications

• Analytical background in research, data science, investigation or audit.
• Stamina for close reading and the discipline to code it consistently.
• Python or SQL experience for working with logs at scale.
• Strong writing skills.

Responsabilités

• Review model trajectories across domains and categorize failures.
• Develop a shared failure taxonomy with dataset leads and researchers.
• Quantify shifts in failure modes across model versions and training runs.
• Propose tasks targeting persistent failures.
• Prepare analyses for customer-facing reports.

Connaissances

Analytical mindset

Close reading

Data analysis

Writing

Outils

Python

SQL

Description du poste

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.