ML Research Engineer, Performance Optimization
Core
Building a platform where AI and automation co-evolve to accelerate scientific discovery by maximizing iteration speed for mid- and post-training runs on 1T+ parameter LLMs.
Role type
ML Research Engineer (Performance Optimization)
Builds
GPU compute and cluster performance optimizations, big data handling pipelines, and RL pipeline configurations for large-scale LLM training.
Domain
Artificial Intelligence / Large Language Models / High-Performance Computing
Deliverable
production ML models
Required skills
LLM inference libraries (vLLM), LLM training libraries (Megatron), multi-node GPU cluster management, reinforcement learning concepts, software engineering excellence, CI/CD pipeline design
Preferred skills
Kernel DSLs (CUDA, CuTe, Triton), GPU networking (NCCL, NIXL), distributed systems (Ray, Flyte), containerized compute backends (Docker, Modal)
Responsibilities
Profile and optimize GPU utilization for training runs, develop kernels for long-context handling and memory optimization, optimize collective/MoE frameworks and RDMA usage, manage mid-training scale data across storage and streaming, optimize async RL cycles and weight sync, establish throughput/accuracy tradeoff standards with researchers, maintain reusable optimized stacks for the team
Seniority
Mid-Senior, hands-on IC
