World Model Data Engineer — Multimodal AI Pipelines

Visa Hunt; Northern (KY); Hybrid; USD 150,000 - 230,000; Full time

Job summary

Runway is hiring Research Engineers to own the data behind our world models. Build datasets, run experiments, and shape data pipelines that enable models to learn across tasks, modalities, and domains.

You’ll work on synthetic data generation, quality control, and large-scale experimentation, influencing models used in creative tools and robotics. Join a team focused on advancing world models and simulation, with opportunities to collaborate across product and creative teams and push the

Qualifications

• 4+ years of experience in machine learning or equivalent
• Experience with large multimodal datasets and generative models
• Ability to design data-centric experiments and analyze results
• Comfort across data analysis, dataset creation, model training and evaluation
• Proficiency with ML frameworks and distributed compute tools

Responsibilities

• Design multimodal, multitask datasets to teach world models new capabilities
• Run controlled training experiments to measure data composition effects
• Build and operate large-scale pipelines for synthetic data generation and filtering
• Define evaluations and benchmarks to track model improvements
• Partner with product and creative teams to align data strategies with target behaviors

Skills

Machine learning

Multimodal datasets

Data-centric approaches

Research stack

ML frameworks

Distributed compute

Tools

PyTorch

JAX

Ray

Kubernetes

Job description

Runway is hiring Research Engineers to own the data behind our world models. Build datasets, run experiments, and shape data pipelines that enable models to learn across tasks, modalities, and domains.

You’ll work on synthetic data generation, quality control, and large-scale experimentation, influencing models used in creative tools and robotics. Join a team focused on advancing world models and simulation, with opportunities to collaborate across product and creative teams and push the