MTS - Research (Kernels & Hardware RL)

Collinear AI•San Francisco, CA

8d•$160,000 - $400,000

About The Position

Collinear is seeking an MTS - Research (Kernels & Hardware RL) team member to build the data engine for frontier AI. The role develops high-fidelity reinforcement learning environments and evaluation stacks that train advanced agents to write efficient GPU kernels and solve hardware optimization problems. The position owns environments end to end, from selecting meaningful tasks and reference implementations to building reliable scoring and preventing reward hacking. The ideal candidate combines technical depth, research intuition, and high agency. They have a strong technical foundation in computer science, electrical or computer engineering, or a related field, or can demonstrate equivalent expertise through significant open-source contributions or industry experience. They bring strong low-level and Python engineering skills, experience building and profiling kernels, understanding of GPU performance and numerical correctness, hands-on experience with LLMs, and a rigorous approach to reproducible performance evaluation. The role collaborates closely with the founders and research staff at a seed-backed startup. The California base salary range is $160,000 to $400,000 per year, depending on experience, skills, and qualifications. The role is also eligible for equity, benefits, and bonuses. Collinear is an equal opportunity employer and provides reasonable accommodations throughout the application and hiring process.

Requirements

A Bachelor’s, Master’s, or PhD in CS, Electrical or Computer Engineering, or a related technical field, or demonstrated proof of work through significant open-source contributions or industry experience.
Strong C/C++ and CUDA or a comparable kernel language, alongside production-quality Python.
Experience building and profiling kernels, with understanding of memory hierarchy, data movement, parallelism, occupancy, and numerical correctness.
Hands-on experience with LLMs and a principled understanding of how they are evaluated and optimized.
Ability to translate a systems problem into an executable task with clear feedback.
Experience conducting reproducible performance experiments, inspecting profiler results, and distinguishing real improvements from measurement noise or benchmark shortcuts.

Nice To Haves

Strong intuition for identifying what matters in complex problem spaces and balancing research exploration with product pragmatism.
Ability to identify gaps, build solutions, and ensure they move real-world metrics for frontier AI labs.
Depth in Triton, CUTLASS, LLVM/MLIR, GPU architectures, vector ISAs, FPGA/HLS, RTL/HDL, EDA tools, or architecture simulators.
Experience building RL environments, agent harnesses, or robust evaluation infrastructure.
A history of shipping high-performance kernels, compiler or hardware tooling, influential systems or ML research, or high-impact open-source projects.
Experience fine-tuning or evaluating large-scale models to deliver frontier performance on open-source benchmarks.

Responsibilities

Design diverse, executable tasks for GPU and accelerator kernels, low-level systems programming, and hardware-aware optimization; extend tasks into FPGA, RTL, or simulation workflows where expertise can create meaningful challenges.
Build reproducible correctness tests and performance graders, benchmark against trusted implementations, and harden execution sandboxes against timing exploits, test leakage, and reward hacking.
Analyze agent trajectories and failure modes, curate high-signal tasks and data, and collaborate on post-training experiments to improve capability across unseen workloads and hardware constraints.
Collaborate daily with the founders and research staff to shape the roadmap and advance AI reliability.
Build reliable compilation, execution, profiling, and evaluation pipelines that support repeated experiments across hardware targets.