About this role
The day-to-day work for this Full Stack Engineer on the RL Data team involves designing and building the tools that researchers and external contributors utilize to create, review, submit, and monitor the environments and tasks used in Cursor’s reinforcement-learning runs. This is a product engineering role situated within a research team, focused on streamlining the process from initial task ideas or data sources to trusted training data. The engineer will own the entire review and acceptance experience, which includes inspecting rollouts, analyzing task-quality signals, and managing workflows that move submissions into training. A key focus is developing authoring interfaces that enable researchers, vendors, and domain experts to quickly and confidently create and refine these environments and tasks.
This role exists to significantly shorten the cycle time between an initial idea and having reliable training data. The team is small and flat, valuing individuals who are truth-seeking, creative, and willing to engage in spirited debate to ship functional code. The ideal candidate thrives in an ambiguous environment, taking broad problems and transforming them into reliable product features by collaborating closely with researchers and domain experts.
What you will do
- Design and build fast, trustworthy workflows for vendors and the research team to interact effectively regarding task creation, iteration, and submission acceptance into training.
- Develop review tools necessary for inspecting and comparing rollouts, transcripts, grader outputs, and other signals related to task quality.
- Create experiences for environment health, failure searching, versioning, and cataloging to ensure training data is easy to understand, manage, and extend.
- Establish a shared component kit to build self-serve interfaces for creating and improving tasks that include inline quality checks.
- Manage the end-to-end experience of task review and acceptance, including rollout inspection and reward-hacking analysis.
What you need
- Demonstrated experience shipping full-stack products, owning systems from the user interface through storage or services, utilizing technologies such as TypeScript and React alongside Node, Python, or Go.
- Experience building dense, data-facing tools, such as transcript viewers, diffing systems, review queues, or operational dashboards, coupled with strong opinions on how structured data should be rendered.
- Background in designing review, QA, moderation, or acceptance workflows where systems must remain honest and prevent users from bypassing checks.
- A strong commitment to data quality and a willingness to inspect raw data.
- Experience with evaluations, graders, reinforcement learning, or data-quality systems is helpful but not mandatory.
- Ability to move quickly under ambiguity and collaborate effectively with researchers and domain experts to turn rough needs into reliable products.
Nice to have
- Experience with evaluations
- Experience with reinforcement learning
- Experience with data-quality systems
Benefits
- San Francisco location
Skills and tech stack
Questions candidates ask
What is the remote policy?
The posting does not explicitly detail the remote work policy for this role.
What is the compensation structure?
The posting does not specify the salary for this position.
What is the technology stack used?
The required technologies include TypeScript, React, Node, Python, or Go.
What is the team structure like?
The organization is described as very flat, and the team is small and talent dense.