Research Scientist, Data

Periodic Labs•Menlo Park, CA

8h•$250,000 - $350,000•Onsite

About The Position

Periodic Labs is an AI and physical sciences company focused on accelerating scientific breakthroughs. The Research Scientist, Data role is central to Scientific AI creation, focusing on evaluations and data. This involves constructing cutting-edge evaluations for advanced scientific use cases, sourcing external datasets, integrating internal experimental data, and building training environments for reinforcement learning. The goal is to ensure the team has the necessary assets in the correct format to evaluate and improve AI models. The role requires collaboration with computational and experimental scientists to translate complex scientific workflows into rigorous evaluations and benchmarks, and partnering with researchers to identify data needs and build the required datasets, environments, and pipelines. The ultimate objective is to establish a strong feedback loop between scientific use cases, model evaluation, and training data.

Requirements

Designed evaluations, benchmarks, or RL environments for language models, agents, or scientific AI systems
Built large-scale data pipelines for LLM pretraining, midtraining, post-training, or evaluation
Strong judgment about dataset and evaluation quality, including scientific relevance, coverage, provenance, licensing, and contamination risks
Strong software and data engineering skills, including familiarity with data processing at scale, dataset versioning, lineage tracking
A research-oriented mindset: you form hypotheses about data, run controlled experiments, measure model outcomes, and iterate with rigor

Responsibilities

Own the evaluation and data strategy across the training stack, identifying capability gaps and shaping the roadmap with leads of physical science and AI research
Work with domain experts to translate advanced scientific workflows into rigorous evals, benchmarks, and RL environments
Source, evaluate, and procure external datasets across chemistry, physics, materials science, mathematics, simulations, and laboratory instrumentation
Build robust pipelines to ingest, clean, and transform for training large-scale datasets from heterogeneous sources
Build tooling and analysis workflows that help researchers inspect data, understand model failures, and determine which evaluations or datasets to develop next