Member of Technical Staff - Post-Training

Focus on alignment tuning, reward modeling, and targeted mitigation for failure modes.

Role Context

This role sits at the boundary between research and production quality. You will convert observed failure claims into concrete post-training objectives, then ship improvements that measurably reduce repeat errors in high-impact workflows.

• Own a subset of high-severity failure modes and mitigation strategies.
• Work directly with evaluation and infrastructure teams to close the train-eval-deploy loop.
• Balance fast iteration with defensible, reproducible experimentation.

First 90 Days

• Map current failure clusters to post-training interventions and define acceptance metrics.
• Ship at least one mitigation cycle with before/after eval evidence.
• Establish a recurring review with product and safety stakeholders on model quality movement.

Cross-Functional Partners

• Model evaluation and red-team teams.
• Inference and data engineering teams.
• Product owners for claims workflow and user-facing quality.

Success Metrics

• Reduction in recurrence rate for prioritized failure modes.
• Improvement in pass rate on internal reliability benchmarks.
• Time-to-mitigation from failure discovery to validated fix.

Responsibilities

• Develop post-training methods to reduce failure rates and improve reliability.
• Design experiments and metrics to evaluate new model behavior.
• Work with engineering and product partners to ship iterative model improvements.
• Communicate insights clearly to technical and product leadership.

Qualifications

• MS/PhD or equivalent experience in ML, data science, statistics, or computer science.
• Experience with model evaluation, experimentation, and applied ML systems.

Compensation

IC4: USD $119,800 - $234,700 (San Francisco: $158,400 - $258,000)

IC5: USD $139,900 - $274,800 (San Francisco: $188,000 - $304,200)