GPU / AI Infrastructure Engineer @ Quant Trading Firm - Hong Kong / Shanghai - 3+ years exp

Tribus; Hong Kong; Sur place; HKD 700 000 - 1 000 000; Plein temps

Résumé du poste

Tribus in Hong Kong is seeking a highly skilled systems engineer to help scale AI and GPU infrastructure for large-scale model training and research.

You will design, build and run GPU clusters, optimize CUDA/kernel performance, and support distributed training platforms used by researchers. The role involves deep kernel and hardware-level debugging on Linux, with access to cutting-edge hardware.

Qualifications

• Strong C++ and/or Python, with hands-on experience in GPU computing, HPC, or low-level systems programming.
• Experience with CUDA, Triton, NCCL, RDMA/InfiniBand, Kubernetes/Slurm, PyTorch internals, distributed training (FSDP/DeepSpeed/Megatron), or inference engines (vLLM/TensorRT-LLM).
• Solid fundamentals in Linux, operating systems, and computer architecture; comfortable debugging performance at the hardware level.
• Open to candidates from AI labs, big tech, cloud/GPU providers, chip companies, HPC centers, or trading firms.

Responsabilités

• Design, build and run large-scale GPU clusters (scheduling, networking, storage, reliability).
• Perform CUDA/kernel optimization, profiling, and tuning across GPU, CPU, memory, and interconnect.
• Develop AI/LLM training and inference platforms used by research teams.
• Support HPC and networking needs using RDMA/InfiniBand, NCCL, and parallel file systems.

Connaissances

C++

Python

GPU computing

HPC

Low-level systems

Linux fundamentals

Formation

Bachelor degree in a relevant field

PhD considered

Outils

CUDA

NCCL

RDMA/InfiniBand

Kubernetes/Slurm

PyTorch internals

TensorRT-LLM

Description du poste

I'm working with a leading global quant trading firm to hire engineers for a new AI and GPU infrastructure team. The firm is scaling up its compute for large-scale model training and research, and needs strong systems engineers who know how to get the most out of the hardware. No finance background needed.

You could be working on one of:

• GPU cluster infrastructure – designing, building and running large-scale GPU clusters (scheduling, networking, storage, reliability)
• Low-level performance engineering – CUDA/kernel optimization, profiling, and tuning across GPU, CPU, memory and interconnect
• AI / LLM platform – training and inference platforms that research teams use day to day: distributed training, model serving, experiment tooling
• HPC & networking – high-throughput systems using RDMA/InfiniBand, NCCL, and parallel file systems at scale

What we're looking for:

• Strong C++ and/or Python, with hands-on experience in GPU computing, HPC, or low-level systems programming
• Experience with one or more of: CUDA, Triton, NCCL, RDMA/InfiniBand, Kubernetes/Slurm, PyTorch internals, distributed training (FSDP/DeepSpeed/Megatron), or inference engines (vLLM/TensorRT-LLM)
• Solid fundamentals in Linux, operating systems, and computer architecture; comfortable debugging performance at the hardware level
• Open to candidates from AI labs, big tech, cloud/GPU providers, chip companies, HPC centers, or trading firms
• ~3-6 years of experience with a stable career history from a strong university, though exceptional PhD candidates with 2-3 years are also considered
• Greenfield build: shape the firm's AI and GPU platform from early stage
• Access to large, cutting-edge GPU compute without big-tech bureaucracy
• Very competitive budget for the right candidateWilling to wait out non-compete periods for strong candidates
• Fast, flexible process with feedback within one week