Multimodal Data Foundations Research Engineer

Madrona Venture Labs; United States; Remote; USD 180,000 - 240,000; Full time

Job summary

Madrona Venture Labs is seeking a Research Engineer to own the data behind our world-modeling efforts. You’ll design datasets, run experiments, and build data pipelines at scale to shape model capabilities for diverse applications from creative tools to robotics.

You will work across data analysis, dataset creation, model training, and evaluation, with a focus on data quality and composition to improve model behavior.

Qualifications

• 4+ years of experience in machine learning.
• Experience with large multimodal datasets and generative models (video, image, or multimodal).
• Deep intuition for how data composition and quality translate to model capabilities.
• Comfort working across the full research stack: data analysis, dataset creation, model training, evaluation, and back again.
• Proficiency with at least one ML framework (e.g. PyTorch, JAX) and distributed compute tools (e.g. Ray, Kubernetes).
• Excitement about building AI that simulates the world.

Responsibilities

• Design multimodal, multitask datasets that teach world models new capabilities and measure their effect on model behavior.
• Run controlled training experiments to understand how data composition drives model performance across tasks and domains.
• Build and operate large-scale pipelines for synthetic data generation, filtering, and quality control.
• Define evaluations and benchmarks that measure improvements in model capabilities.
• Partner with product and creative teams to translate target behaviors into concrete data strategies.

Skills

Machine learning

Multimodal datasets

Model training & evaluation

Data-centric thinking

Distributed computing

Data analysis

Tools

PyTorch

JAX

Ray

Kubernetes

Job description

Madrona Venture Labs is seeking a Research Engineer to own the data behind our world-modeling efforts. You’ll design datasets, run experiments, and build data pipelines at scale to shape model capabilities for diverse applications from creative tools to robotics.

You will work across data analysis, dataset creation, model training, and evaluation, with a focus on data quality and composition to improve model behavior.