Research Scientist/Research Engineer, Midtraining

periodic-labsMenlo Park, CAEngineering

Apply on company site

Opens the official periodic-labs careers page

Salary

$250K–$350K

Job type

Full-time

Experience

Mid

Posted

30 Sept 2026

Status

Open · checked 7d ago

About this role

This role suits an applied researcher or engineer who wants to improve scientific reasoning in frontier models and is based near Menlo Park, California. Periodic-labs seeks a full-time Research Scientist or Research Engineer for midtraining to help build advanced models aimed at scientific discovery.

Role Overview

Periodic-labs is hiring a Research Scientist or Research Engineer to focus on midtraining in Menlo Park, CA. The position involves taking base models and enhancing their scientific reasoning capabilities through data curation, synthetic data generation, evaluation building, and large-scale training experiments.

Key Responsibilities

Find, handle, and curate new scientific data sources for training large-scale models.

Create high-quality synthetic data to address gaps in scientific knowledge and reasoning.

Construct evaluations that align with downstream scientific task performance in collaboration with physicists, chemists, and reinforcement learning researchers.

Apply methods such as on-policy distillation and self-distillation to enhance model capabilities.

Plan and execute large-scale training experiments alongside supercompute engineers for efficient scaling across thousands of GPUs.

Create internal tools to analyze how data decisions affect model intelligence.

Requirements

Background in training large language models on curated datasets containing trillions of tokens.

Background on a dedicated evaluations team supporting major production training runs.

Practical application of on-policy distillation, self-distillation, or comparable methods within active training pipelines.

Familiarity with compute-optimal hyperparameters and scaling laws.

Ability to operate across training infrastructure, evaluations, and data.

Nice to Have

Background optimizing reliability and throughput for distributed training runs at scale.

Familiarity with AI for science or training using specialized scientific domain data such as materials or proteins.

Background developing synthetic data or evaluations for unverified tasks and monitoring performance during live runs.

Skills and Tools

Large Language Models

Distributed Training

Data Curation

Synthetic Data Generation

Model Evaluations

Self-Distillation and On-Policy Distillation

Scaling Laws

Supercompute Infrastructure

Salary and Benefits

Compensation range of $250,000 to $350,000

Equity

Visa sponsorship and relocation assistance support

About the Company

Periodic-labs is an artificial intelligence and physical sciences organization constructing state-of-the-art models designed to accelerate breakthroughs in energy, materials, and related fields. The enterprise receives backing from investors and experiences rapid growth.

How to Apply

Interested candidates should apply directly on the official periodic-labs careers page.

Skills and tech stack

LLMs Distributed Training Infrastructure

Questions candidates ask

Where is this job located?

This position is located in Menlo Park, CA.

What is the salary range for this role?

The compensation ranges from $250,000 to $350,000 plus equity.

Does the company provide visa sponsorship?

Yes, visa sponsorship is available and the company provides assistance with the process.

What is the minimum education requirement?

The minimum requirement is a Bachelor's degree or similar experience.

What is the primary focus of the midtraining role?

The focus is on improving base model scientific reasoning through data curation, synthetic data creation, evaluation building, and large-scale training experiments.

Ready to apply?

The application is on the official periodic-labs careers page.

Apply on company site