ML Compiler and Systems

Department: Engineering & Research

Location: Bay Area

Employment Type: FullTime

What You'll Do

• Lead the evolution of our high-performance robotics simulation platform

• Design and implement the compute infrastructure and data flow mechanisms to optimize performance for physics simulation and foundation model training

• Lead development of our compiler stack, focusing on JIT compilation, LLVM IR, and GPU codegen to minimize compile time and maximize runtime performance

• Collaborate with the team to improve the compiler's support for differentiable programming, crucial for training neural networks within simulations

• Stay current on state-of-the-art ML compilers—such as those in torch, Triton, and JAX—and decide which techniques and approaches are best suited for our application

• Work closely with simulation and robotics engineers to align compiler enhancements with application needs

• Contribute to relevant open-source projects and participate actively in the broader compiler and systems community

What You’ll Bring

• Strong background in compiler construction, particularly in JIT compilation and LLVM-based code generation

• Extensive experience with GPU programming models (e.g., CUDA, Vulkan) and understanding of GPU architecture

• Track record as a core contributor to GPU programming infrastructure—such as Torch, JAX, Mojo, Taichi, or Warp

• Proven ability to profile and optimize complex systems for performance and scalability

• Understanding of automatic differentiation and its application in simulation and machine learning contexts

• Excellent communication skills and a collaborative approach to problem-solving

• Enthusiasm for contributing to and engaging with open-source communities