ML Researcher Model Adaption / Inference Optimization

A well-resourced AI research and infrastructure team building models that keep learning after deployment, along with the inference systems that serve them at frontier-level speed and cost. Small, technically deep team; distributed by default with regular in-person collaboration.

Requirements

Strong first-principles reasoning and deep technical expertise (e.g. ML systems, compilers, mathematics, physics, competitive programming, security, or HPC)

Hands-on experience with GPU-level performance engineering (Triton/CUDA) and/or model training and post-training systems

Creative, high-agency builder who has mastered something hard, with a track record of tackling genuinely novel technical problems

Comfortable moving fluidly between open-ended research and shipping production code

Responsibilities

Design and optimize inference systems that serve continuously adapting models with frontier-level performance and economics

Develop hardware-informed quantization approaches and write/optimize Triton and CUDA kernels

Convert agent traces into synthetic reinforcement learning environments for continued model training

Identify and resolve performance bottlenecks across compute, memory, and communication

Efficiently deploy LoRAs, learned memory, specialized weights, and other adaptation techniques in production

Implement policy updates derived from live model interactions and post-deployment feedback loops