Helix AI Engineer, Modeling

About the role

Figure is constructing a new class of autonomous humanoid robots intended to operate within human environments, and the Helix team is responsible for the core intelligence that allows these machines to see, think, and act. We are looking for an engineer who will own the design and implementation of the foundational AI models that power perception, cognition, and motion. In this position, you will translate high-level robotic goals into concrete model architectures and training strategies. You will own the technical decisions that shape how these systems understand the world and generate actions. The role requires deep collaboration with researchers and builders across multiple specialized domains to ensure the models are both cutting edge and deployable. You will be responsible for turning ambiguous problems in embodied intelligence into well-defined engineering challenges. Ultimately, your work will directly determine the capabilities and reliability of the robots we build.

Key facts

What you'll do

• Design and implement model architectures capable of ingesting and interpreting multimodal inputs such as vision, language, and proprioceptive signals.
• Engineer systems that construct structured representations of objects, environmental dynamics, and physical interactions to support reasoning.
• Advance multimodal learning techniques by improving cross-modal reasoning, alignment, and fusion methods.
• Optimize model performance with a focus on generalization across tasks, long-horizon reasoning, and operational reliability.
• Oversee the complete model lifecycle, guiding concepts from early research and prototyping through rigorous training and final deployment.
• Partner with specialized teams in pretraining, video, generative models, reinforcement learning, and robot learning to integrate novel architectures.
• Develop evaluation frameworks and run experiments that analyze model behavior, using findings to drive iterative improvements.
• Invent new modeling paradigms specifically tailored to the constraints and opportunities of embodied AI.
• Define the standards and best practices that shape how models are built, validated, and monitored in production.
• Translate high-level product and research objectives into concrete model requirements and success metrics.
• Identify failure modes and design architectural or training interventions that mitigate those risks.
• Contribute to the broader research agenda by documenting approaches and sharing insights across the Helix organization.

Requirements

• Possess professional experience designing and training deep learning models intended for language, vision, or multimodal systems.
• Demonstrate deep knowledge of modern architectures, with a specific focus on transformers and related technologies.
• Show a proven ability to improve model outcomes through systematic architectural experimentation.
• Exhibit proficiency in Python and PyTorch for building and training complex models.
• Display strong experimental discipline and the patience required to refine model designs over extended iterations.
• Apply software engineering skills to build maintainable, reliable, and scalable systems.
• Work autonomously on complex technical challenges that are often ambiguous and poorly defined.
• Communicate findings and trade-offs clearly to both technical and non-technical stakeholders.

Nice to have

• Have hands-on experience with vision-language-action models or other advanced multimodal systems.
• Bring a background in world models, representation learning, or structured prediction.
• Have previous work on frontier models at organizations such as Google DeepMind, OpenAI, Meta, Anthropic, or xAI.
• Possess familiarity with robotics, embodied AI, or real-world machine learning deployments.
• Have experience with distributed systems or large-scale training infrastructure.
• Maintain a track record of publications in fields such as NLP, computer vision, multimodal AI, or machine learning.

Skills & tools

• Python
• PyTorch
• Transformer architectures
• Multimodal learning
• Deep learning

Practical notes

• Base salary range: $200,000 - $400,000.
• Final compensation packages are determined by individual experience, skills, and knowledge, and will be disclosed upon receiving an offer.
• The role requires full-time, in-office presence for five days each week in San Jose, Canada.
• Candidates must be eligible to work in the country where the position is located without sponsorship for this role.
• This position reports to the Helix organizational structure and collaborates closely with cross-functional robotics and AI teams.
• The views and ideas expressed during the application process will be used to assess technical fit for modeling challenges.