Reflection - Member of Technical Staff - Research Software Engineer

• Designing and optimizing large-scale training loops and data pipelines. • Implementing state-of-the-art techniques and ensuring they are numerically stable and computationally efficient. • Building internal tooling for launching, monitoring, and reproducing complex experiments. • Diagnosing deep bottlenecks across the training stack (GPU memory issues, communication overhead, dataloader stalls). • Translating research prototypes into reusable, production-grade infrastructure. • What You'll Work With • Distributed Training • GPU parallelism (data, tensor, pipeline, expert) • Large-scale distributed training infrastructure • Communication optimization (NCCL, RDMA, GPU interconnects) • FSDP / ZeRO and model sharding • Orchestration & Runtime Systems • Ray, Kubernetes, Slurm • Distributed runtimes and async systems • Containerization and sandboxing • Megatron-style training stacks • Triton / custom kernels • Data Infrastructure • Large-scale dataset curation pipelines • Deduplication and filtering systems • Tokenization and preprocessing • Distributed data processing frameworks