Research Engineer, Domain Scaling

About the role

The Domain Scaling team has the goal to make Claude world-class at real-world knowledge work in domains like finance, healthcare, and legal. This is a unique role that combines executing directly on applied research and data sourcing (real-world and synthetic) to improve our models. You'll own the end-to-end process of creating RL environments for new capabilities: identifying high-value tasks, designing reward signals, managing vendor relationships, and measuring impact on model performance.

Responsibilities

• Own the data strategy for knowledge work verticals end-to-end, from task sourcing through RL training

• Manage technical relationships with external data vendors, including evaluation of data quality and reward design

• Collaborate with domain experts to design data pipelines and evaluations

• Explore novel ways of creating RL envs for high value tasks

• Develop and improve QA frameworks to catch reward hacking and ensure env quality

• Run generalization experiments to measure how data strategy changes improve model capabilities

• Partner with other RL research teams and product teams to translate capability goals into training envs and evals

You may be a good fit if you

• Have experience with fine-tuning large language models for specific domains or real-world use cases

• Have experience with reinforcement learning, reward design, or training data curation for LLMs

• Are comfortable managing technical vendor relationships and iterating quickly on feedback

• Find value in reading through datasets to understand them and spot issues

• Have strong cross-functional collaboration skills

• Are passionate about making AI more useful and accessible across different industries

• Are excited about a role that includes a combination of applied research and hands-on data work

Strong candidates may also

• Have experience training production ML systems

• Have experience designing evals or benchmarks for LLMs

• Have domain expertise in a vertical where we would like to make our models more useful

• Have experience working with external vendors or technical partners