1 month ago
Responsibilities
- Design end-to-end synthetic data recipes that generate large, diverse, verifiable task sets from traces, documents, codebases, and workflows.
- Build scalable data-generation pipelines, agent systems, verifiers, and anti-hacking mechanisms.
- Control task difficulty and diversity so datasets remain learnable, realistic, and resistant to template collapse.
- Run targeted ablations and analyses to establish that datasets improve model capabilities and to rule out grader or data artifacts.
- Use traces and training results to address capability gaps and convert successful recipes into reusable infrastructure.
- Partner with research and RL Data teams to translate model goals into synthetic data plans.
Requirements
- Research experience with synthetic or post-training data in industry or academia, including designing a generation method, shipping data into training, and demonstrating model impact.
- Strong software engineering fundamentals and experience building non-trivial data pipelines, agent systems, or evaluation frameworks.
- Ability to design training tasks that teach meaningful capabilities, can be graded reliably, and resist reward hacking.
- Careful, systematic analysis skills spanning aggregate metrics and close reading of raw outputs.
- Ability to turn open-ended problems into repeatable, high-quality data production systems.
Categories
About Cursor
Cursor is an AI-powered code editor and coding assistant from Anysphere, built for professional software engineers and enterprise teams. It combines a VS Code–style editor with integrated AI chat, code generation, and repo-aware refactoring, and is sold via individual, team, and enterprise subscriptions. The company also targets sectors like healthcare and state and local education with field engineers who support evaluations and proofs of concept.