Member of Technical Staff - Low Level & Kernels Capabilities

About the Company

Preference Model is building automated ML research engineering. The founding team has previous experience on Anthropic’s data team building data infrastructure and datasets behind Claude. We are partnering with leading AI labs to push AI closer to achieving its transformative potential.

Responsibilities

• Design and build low level / kernel-focused reinforcement learning (RL) environments that target a specified model and difficulty distribution.
• Choose which environments are worth building, targeting niche domains and real hardware features like tiling, streaming, and vector ISAs.
• Work with interesting hardware or simulators such as FPGAs, novel accelerators, or gem5.
• Build correctness and performance scoring that is deterministic and resistant to reward hacking.
• Ensure tasks are research-motivated and grounded in benchmarks where models currently lag.

Requirements

• Strong low-level/systems engineering fluency in C, C++, or CUDA.
• Ability to work with assembly when necessary.
• Strong, engineering-quality Python for production code, automation, and data analysis.
• Hardware-aware coding mindset considering memory hierarchy, occupancy, and parallelism.
• Experience in kernel development and iterative optimization using profilers.
• An adversarial mindset to create ungameable scoring systems.
• Hands-on experience with LLMs.
• Ability to work with high autonomy and ownership.

Preferred Qualifications

• Experience shipping kernels that approach SOTA performance.
• Depth in niche hardware targets or ISAs (FPGA/HLS, RISC-V Vector, DSPs, SIMD/AVX, TPUs).
• Experience in HPC, hardware design (RTL/HDL), compilers (MLIR/LLVM, Triton), or formal verification.
• Ability to translate architecture papers into running code.
• Open-source contributions.
• Competitive programming background in low-level languages.
• Experience building RL environments or agent harnesses.