Research Team Lead – Distributed AI Systems & Large-Scale Infrastructure

Requirements

• B.Sc. or higher in Computer Science, Computer Engineering, Electrical Engineering, or a closely related field

• 8+ years of experience in systems software, distributed computing, or AI infrastructure, with 3+ years in a leadership or team lead role

• Deep expertise in large-scale communication systems: collective communication, RDMA, network topology-aware routing, and bandwidth optimization

• Hands-on experience building software infrastructure for distributed training on custom accelerators or heterogeneous hardware (GPU, NPU, TPU)

• Strong knowledge of runtime systems: scheduling, execution graphs, kernel dispatch, synchronization primitives, and pipeline management

• Experience with memory management at scale: activation checkpointing, tensor offloading, rematerialization, KV cache management

• Proficiency in C/C++ and Python, with a focus on high-performance, production-quality code in Linux environments

• Proven ability to define technical vision, lead multi-person projects end-to-end, and deliver results under research and engineering timelines

• Excellent communication skills in English — confident presenting to international audiences, writing technical reports, and driving cross-team alignment

• Strong collaborative mindset and experience working in globally distributed, multicultural teams

Ways to Stand Out From the Crowd

M.Sc. or Ph.D. in a relevant field, with a strong publication record at systems or ML venues (EuroSys, OSDI, SC, NeurIPS, MLSys, ISCA)

• Hands-on experience with communication frameworks such as NCCL, MPI, HCCL, or UCX

• Experience with compiler and graph optimization for AI workloads (XLA, TVM, Triton, or custom operator fusion)

• Background in mixed-precision training, model parallelism (Tensor Parallelism, Pipeline Parallelism, Expert Parallelism), and large model co-design

• Experience profiling and debugging performance bottlenecks on heterogeneous clusters using tools like Chrome tracing, nsight, or custom profilers