Research Engineer, Search and Knowledge Post-Training

Role Focus:

• Advance the science and engineering to make Claude a trustworthy searcher by defining hypotheses and designing experiments.
• Turn search post-training from a craft into a measurable science with cleanly isolated variables and reproducible signal.
• This work sits at the intersection of reinforcement learning, retrieval, and evaluation, shaping Claude's behavior in evidence-based settings.

Key Responsibilities:

• Own research direction end-to-end, from hypothesis formation to experiment design and training runs.
• Build controlled experiment infrastructure to study environmental factors and design evaluations that distinguish genuine reasoning.
• Drive optimization rigor through efficient experiment design and ablations, and set the team's experimental standards.
• Collaborate with researchers across post-training, RL infrastructure, and product to translate model behavior into training signals.

Qualifications:

• Must have an unusually rigorous, quantitative mindset and be an outstanding software engineer in Python.
• Must have shipped real ML research repeatedly with a taste for worthwhile experiments and operate well with high autonomy.
• Preferred experience includes hands-on RL with LLMs, background in search/retrieval/RAG, and experience in research-heavy environments.
• Prior published research on LLMs, RL, retrieval, or calibration is a plus, as is experience with distributed training systems.