Research Scientist, Agentic Data & Benchmarking

Key responsibilities

Benchmarking & evaluation 

• Design and run evaluations of agentic capabilities — multi-step reasoning, tool use, long-horizon planning, computer use, and safety properties — turning ambiguous notions of "intelligence" into defensible, reproducible metrics. 

• Build and harden evaluation harnesses so benchmarks run reliably at scale against training checkpoints, with clear signal on regressions and model health. 

• Run experiments characterizing how prompting, sampling, scaffolding, and environment design affect agentic performance on internal and public benchmarks. 

• Diagnose anomalous eval results mid-training-run — determine whether the cause is the model, the data, the harness, or the infrastructure — and communicate the answer clearly. 

Agentic data 

• Source, generate, and curate high-quality agentic training data: trajectories, tool-use traces, and task datasets for new capabilities. 

• Design and scale RL environments and reward signals, and measure their impact on model performance. 

• Manage technical relationships with external data vendors and domain experts, evaluating data quality and iterating quickly on feedback. 

• Develop QA frameworks that catch reward hacking, label noise, and contamination, keeping data and benchmark quality high. 

Across both 

• Contribute to technical reports, research publications, and open-source benchmarks and tooling. 

• Partner with research and product teams to translate capability goals into measurable data and evaluation artifacts. 

Qualifications

Academic qualifications 

• BS, MS, or PhD (or equivalent experience) in Computer Science, Machine Learning, or a related field. 

Minimum qualifications 

• 2+ years of experience with a clear emphasis on evaluations and/or training-data curation for ML systems (related areas: LLM training/fine-tuning, RL, or distributed ML systems). 

• Strong Python and PyTorch development experience. 

• Demonstrated experience designing and deep-diving into evaluations, or curating and generating training datasets — ideally both. 

• Hands-on experience using LLM agents in your personal or professional work. 

• A habit of reading through raw data and trajectories to understand them and spot issues, and an instinct to distrust a metric until it's validated. 

Preferred qualifications 

• Experience with reinforcement learning, reward design, or RL environment construction for LLMs. 

• Background in statistics and experimental design — a feel for signal-to-noise, statistical power, and contamination in evaluations. 

• Experience with large-scale dataset sourcing, curation, and processing, including working with external vendors or domain experts. 

• Strong knowledge of the literature on agent evaluation, RL, LLM reasoning, and tool use. 

• Experience building or operating data pipelines and evaluation infrastructure reliable at scale (e.g., PyTorch, Ray). 

• Experience evaluating or generating data for software-engineering or computer-use agents. 

• Contributions to published research, public benchmarks, and/or open-source ML software. 

Representative projects

• Stand up a new agentic benchmark from scratch — define the task, build the dataset and scoring, validate against known signals, and ship a view that makes the result legible to researchers and leadership. 

• Build an RL environment for a new high-value capability: design the reward, generate and QA the trajectory data, and measure the lift on model performance. 

• Diagnose a mid-training regression: an eval suite returns anomalous numbers and you determine whether it's the model, the harness, the data, or the infrastructure. 

• Partner with an external data vendor or domain expert to source high-quality trajectories, then build the QA framework that keeps reward hacking and contamination out. 

• Take a flaky distributed eval pipeline and make it reliable — better retries, better observability, faster feedback to researchers.