GPU Kernel Engineer

TypeSafe AI•San Francisco, CA

2d•Onsite

About The Position

TypeSafe is a frontier model lab building reliable and general AI systems to power economically valuable automation. Their mission is to usher in a new era of Transformative Artificial Intelligence (TAI). They are rethinking the LLM stack from first principles, building a new kind of general frontier model designed for real-world reliability, decision-making, and autonomy in production. The company is a small, fast-moving team from OpenAI, Google Brain, and Meta/FAIR, backed by top-tier investors. Since mid-2024, they have been engineering the foundation for what comes after the current "state-of-the-art" — a model that actually gets things done. This role is for a GPU kernel engineer with deep, low-level CUDA expertise to make their training and inference faster and more efficient. The engineer will write and optimize custom kernels, profile and eliminate bottlenecks, and work close to the metal across their model stack.

Requirements

Deep CUDA / GPU kernel expertise and a track record of real performance wins
Have built and optimized inference / training kernels
Hands-on LLM training experience (real, not at a hobbyist level)
Reason from first principles about performance, memory, and parallelism
Responsible, ownership-inclined team players who are mission aligned and excited to go all-in

Responsibilities

Write, optimize, and maintain high-performance GPU kernels (e.g., in CUDA / CuTe DSL) for training and inference
Profile end-to-end performance and eliminate bottlenecks across the stack
Partner with research and platform engineers to squeeze maximum throughput and minimum latency out of our hardware