Job Title: MLOps Engineer (PyTorch)

Location: Singapore

Job Type: Full-time

About the Opportunity

Our client is seeking an MLOps Engineer with a strong background in systems programming and infrastructure engineering. This role is focused on owning and evolving the on-premise infrastructure that powers their advanced PyTorch-based training workloads.

This position is a perfect fit for an engineer who is not just focused on model outcomes, but on the quality and robustness of the underlying systems. You will be responsible for building high-quality, maintainable training pipelines, solving low-level systems and networking challenges, and ensuring the training codebase is clean, scalable, and built to last.

Key Responsibilities

• Architect, build, and maintain end-to-end training and inference pipelines using PyTorch.
• Develop and maintain high-quality, robust tooling in both Python and C++ to support the entire model training lifecycle.
• Take full ownership of the core training codebase, enforcing best practices for clarity, modularity, reproducibility, and performance.
• Design and implement workflows for checkpointing, resuming jobs, model versioning, and experiment tracking.
• Proactively optimize compute workloads for bare-metal environments, focusing on I/O bottlenecks, CPU/GPU utilization, and memory efficiency.
• Troubleshoot and debug complex, low-level issues, including networking bottlenecks, distributed training errors (e.g., NCCL), and hardware faults.
• Configure and manage all ML environments, including containers, package management, GPU drivers, and runtime configurations.
• Monitor and debug large-scale training jobs running across multiple nodes and GPUs.

Required Qualifications (You Should Have)

• Deep, expert-level knowledge of PyTorch, including DDP (DistributedDataParallel), mixed precision training, and TorchScript.
• Advanced programming skills in both C++ and Python.
• A solid background in computer science fundamentals (data structures, algorithms, concurrency, operating systems).
• Hands-on experience debugging and tuning bare-metal servers, including Linux administration, kernel parameter tuning, and BIOS tuning.
• A strong understanding of low-level networking (e.g., RoCE, InfiniBand), interconnects, and distributed training protocols like NCCL and MPI.
• A proven track record of building reliable, reproducible pipelines for both model training and evaluation.
• Experience with job schedulers (e.g., SLURM, or custom runners) and cluster monitoring tools.

Preferred Qualifications (Nice-to-Have)

• Experience with non-standard deployments, such as on-premise local clusters or edge devices (i.e., not public cloud).
• Active contributions to PyTorch or other open-source ML/HPC tools.
• Familiarity with Infrastructure-as-Code (IaC) tools like Ansible, Terraform, or Nix.
• Experience building out a full logging, observability, and alerting stack for training workloads.

How to Apply

Interested candidates are invited to submit their resume, detailing their experience in managing PyTorch workloads on bare-metal infrastructure.