AI Engineer - Robotics

Jobzhr; San Francisco (CA); On-site; USD 150,000 - 190,000; Full time

Job summary

Jobzhr is seeking an AI Engineer to help build a vision-language-action foundation model that operates on real robots. The role focuses on turning perception, reasoning, and motor control into robust, real-time decisions on embedded edge hardware.

You will join a small founding team drawn from robotics, autonomous systems, and foundation-model research, moving concepts from prototype to production-grade deployments on physical platforms.

Qualifications

• 2+ years hands-on AI/ML modeling experience (senior candidates: 5+ with leadership).
• Strong background in reinforcement learning, imitation learning, transformers, multimodal architectures.
• Exposure to embodied AI or robotics is a major plus.
• Proficiency in Python and PyTorch; CUDA a bonus.
• MS or PhD preferred, ideally from a strong program; side projects and portfolios welcomed.
• Must be eligible for U.S. export-controlled work (U.S. person status - citizen or green card - required).

Responsibilities

• Building vision-language-action models that fuse perception, reasoning, and control.
• Multimodal model fusion, continual learning, and domain adaptation.
• Real-time inference on embedded / edge systems (GPU, sub-200ms latency).
• Dynamic memory systems and multi-agent coordination (formations, convoys, swarms).
• Long-horizon autonomy and chain-of-thought reasoning for multi-step missions.
• Taking research concepts through to production-grade, deployed systems.

Skills

Python

PyTorch

Reinforcement Learning

Imitation Learning

Transformers

Multimodal Architectures

Production-grade Code

Education

MS or PhD preferred

Tools

CUDA

JAX

TensorFlow

Job description

AI Engineer - Vision-Language-Action Foundation Models

Location: On-site, Bay Area (CA)

About the role

A well-funded, early-stage physical AI company is building a vision-language-action (VLA) foundation model that turns real-world robots into intelligent, autonomous agents. Think of it as a large multimodal model that outputs motor control instead of text - perceiving the environment, reasoning about it, and acting on it in real time.

The model runs onboard, at the edge, in real time, even in GPS- and comms-denied environments. Prototypes across ground and air platforms are already operational in the field today. You'd be joining a small, hand-picked founding team drawn from leading robotics, autonomous-vehicle, and foundation-model backgrounds - solving genuinely unsolved problems in embodied intelligence, and shipping them onto real hardware rather than benchmarks.

This is applied AI, not pure research. If you want your models controlling physical machines in the real world, this is that role.

What you'll work on

• Building vision-language-action models that fuse perception, reasoning, and control
• Multimodal model fusion, continual learning, and domain adaptation
• Real-time inference on embedded / edge systems (GPU, sub-200ms latency)
• Dynamic memory systems and multi-agent coordination (formations, convoys, swarms)
• Long-horizon autonomy and chain-of-thought reasoning for multi-step missions
• Taking research concepts through to production-grade, deployed systems

What we're looking for

• 2+ years hands-on AI/ML modeling experience (senior candidates: 5+ with demonstrated leadership)
• Strong background in one or more of: reinforcement learning, imitation learning, transformers, multimodal architectures
• Exposure to embodied AI or robotics is a major plus
• Proficiency in Python and PyTorch (JAX, TensorFlow, CUDA a bonus)
• Proven ability to translate research into robust, production-level code
• MS or PhD preferred, ideally from a strong program; side projects and portfolios welcomed
• Must be eligible for U.S. export-controlled work (U.S. person status - citizen or green card - required)

Who thrives here

• Comfortable in a fast-moving, high-autonomy startup environment
• Deeply curious - keeps current with the latest VLM / RL / CV literature
• Biased toward execution, but grounded in strong research fundamentals
• Motivated by hard real-world problems and hands-on work close to the hardware