Research Engineer, Training & Inference
Core
Building and optimizing the full-stack reinforcement learning infrastructure, from low-level simulators and custom communication primitives to distributed training loops and inference engines.
Role type
Senior IC reinforcement learning systems engineer (full-stack)
Builds
Proprietary RL training and serving infrastructure for AI agents
Domain
Artificial Intelligence / Reinforcement Learning / High-Performance Computing
Deliverable
production ML models
Required skills
Python, PyTorch/JAX/TensorFlow, distributed training, collective communication primitives (NCCL), GPU-accelerated cloud infrastructure, CUDA kernels
Preferred skills
C++, Triton/CuTeDSL/TileLang/CUDA/CUTLASS/ThunderKittens, FSDP/Tensor Parallelism, asynchronous data sampling
Technologies
Python, PyTorch, JAX, TensorFlow, CUDA, NCCL, Triton, CuTeDSL, TileLang, CUTLASS, ThunderKittens, FSDP
Responsibilities
Maintain and optimize proprietary RL training and serving infrastructure from Python API to CUDA kernels; Maximize throughput of RL systems from data generation to model training using sharded multi-node training; Optimize inference stack for high-throughput RL and low-latency LLM traffic; Identify and resolve performance bottlenecks in distributed clusters for multi-billion parameter models.
Seniority
Senior, hands-on IC