Research Engineer, Post-Training Data

NxT Level•San Francisco, CA

23h•Onsite

About The Position

Our client is building the next generation of post-training data infrastructure for frontier AI labs. The company’s core belief is that research and data production are inseparable. The best post-training data is not created through brute-force labeling or headcount alone — it requires domain expertise, research judgment, ML fluency, and the ability to build systems that scale data generation superlinearly. Inbound demand from frontier AI customers is growing faster than the current team can support, and our client is hiring early technical talent to help meet that demand. This is an opportunity to join a team building high-quality post-training data, reinforcement learning environments, long-horizon tasks, and domain-specific evaluation workflows for some of the most advanced AI systems in the world. Our client is hiring a Research Engineer, Post-Training Data to own the full lifecycle of post-training model and data work. This role blends AI research, ML engineering, software engineering, and data production into one function. The ideal candidate is not just a researcher and not just an engineer — they are someone who can understand a domain deeply, identify what makes a task realistic and economically valuable, build the environment or data-generation system, and improve the model through high-quality post-training data. The company is indexing heavily on research and ML horsepower, strong software ability, high slope, and data taste.

Requirements

Genuine spike in ML, AI research, or a technical research domain
Experience or strong interest in post-training, reinforcement learning, evaluations, interpretability, or related areas
Strong software engineering ability and comfort building production-quality tools or research systems
Ability to post-train models and build or curate high-quality post-training data
Strong judgment around data quality, task design, and what makes a problem valuable to AI labs
Fast problem-solving ability and strong code comprehension
Ability to work across unfamiliar domains and ramp quickly
High-slope learning profile with strong technical curiosity
Comfort operating in ambiguous, research-heavy environments
Evidence of deep commitment, strong output, and public artifacts or meaningful technical work

Nice To Haves

Domain or research expertise in a field such as neuroscience, physics, chemistry, systems, chip design, science, or another technical discipline
Strong ML or software engineering core
Experience with PyTorch, ML pipelines, model debugging, or research tooling
Experience building environments, evaluations, or data-generation systems
Exposure to post-training, RL, long-horizon tasks, or frontier model workflows
Experience at frontier AI labs, AI infrastructure companies, or high-talent technical teams
Experience with post-training data, RL environments, evals, model behavior, or interpretability
Experience building data products, research tooling, or developer workflows for AI teams
Experience in domains like chip design, physics, chemistry, biology, systems, or scientific computing
Public artifacts, research projects, open-source work, technical writing, or demos that show exceptional ability
Experience creating tasks or environments that require long-horizon reasoning
Experience scaling data generation through automation rather than manual labor

Responsibilities

Own post-training workflows end to end, from infrastructure provisioning through data generation and curation
Build and curate high-quality post-training data from model and environment generation processes
Produce reinforcement learning environments and long-horizon tasks across complex domains
Work on research-loop, science, chip, physics, and other technically deep task environments
Build systems that scale data generation superlinearly through self-improving developer processes
Improve data generation through better tools, workflows, automation, evaluation, and model feedback loops
Exercise and develop strong data taste around what problems matter in a given domain
Identify realistic, economically valuable tasks and environments that frontier AI labs will actually want
Blend research and data production into a single technical function
Debug ML pipelines, understand unfamiliar code quickly, and solve open-ended technical problems
Help define the standards for what high-quality post-training data should look like across domains