Research Scientist

Pantograph; San Francisco (CA); On-site; USD 150,000 - 230,000; Full time

Job summary

Pantograph is training general models starting from large-scale video and expanding to robotics. We seek research scientists to scale simple methods across massive datasets. You’ll work on pre-training, self-supervised RL, and video modeling, with a focus on scalable, interpretable metrics and rigorous evaluation.

We value contributions from open-source projects and lab experience, and welcome researchers who enjoy translating theory into practical, real-world robotic systems.

Qualifications

• Experience with large-scale pre-training (video, multimodal, image, or language).
• Experience with self-supervised, goal-conditioned, or unsupervised RL.
• Robotics models trained on large-scale data.
• Video generation or sequence modeling over high-dimensional observations (pixels).
• Trained models on large GPU clusters and are comfortable working with Kubernetes.
• Believe simple scalable methods beat complexity; aim for simplest effective solution.
• Strive to define expressive metrics and measure them accurately.
• Value scientific integrity and understand true effects of interventions.
• Nice to have: Experience with JAX; interest in adjacent problems; strong background in measure-theoretic probability, stochastic processes, optimization.

Skills

Large-scale pre-training

Self-supervised RL

Robotics data models

Video sequence modeling

GPU clusters

Kubernetes

Simple methods

Expressive metrics

Tools

JAX

Job description

Pantograph is training general models that start by watching internet-scale video and end up on robots. We think the path to capable robots runs through general intelligence rather than narrow, robot-specific skills. We're scaling simple methods across video games, real-world video, and our own fleet of affordable, durable robots.

We're looking for research scientists who want to scale simple methods across the largest datasets available.

You might be a good fit if you:

• Have experience with one or more of:

  • Large-scale pre-training (video, multimodal, image, or language)

  • Self-supervised, goal-conditioned, or unsupervised RL

  • Robotics models, especially those trained on large-scale data

  • Video generation or other large-scale sequence modeling over high-dimensional observations (e.g. pixels)

• Have trained models on large GPU clusters and are comfortable working with Kubernetes

• Believe simple methods that scale beat complicated ones that don't, and reach for the simplest thing that could work

• Strive to find simple, expressive metrics and measure them accurately

• Value scientific integrity and seek to understand the true effect of different interventions

Nice to have:

• Experience with JAX

• Interest in problems adjacent to the critical path — new modalities, alternatives to text for reasoning, pixel-space modeling, or automating research itself

• A strong background in proof-based mathematics, including topics such as:

  • Measure-theoretic probability

  • Stochastic processes

  • Optimization theory

We care much more about what you can do than any specific credential. We're interested in published work or lab experience, but equally in strong open-source contributions or personal projects. If you're excited about scaling general models that learn from and act in the real world, we'd love to talk.