AI Engineer, AI & Applications

Location: Singapore or Australia, specifically Launceston, Tasmania or Sydney, New South Wales

Company

hirify.global develops sustainable AI infrastructure and production-grade systems for hyperscale model training and AI applications.

What you will do

• Build production-ready TorchTitan and Megatron-LM training recipes, including model configurations, parallelism strategies, and checkpointing patterns.
• Document tuning parameters and expected throughput for distributed training workloads at different scales.
• Create and validate multi-node NCCL communication patterns on AI Factory Kubernetes and Slurm clusters.
• Design benchmarking suites covering accuracy, latency, throughput, cost per token, energy efficiency, and MFU.
• Implement offline evaluation harnesses, leaderboard tracking, and fine-tuning experiments using LoRA and QLoRA.
• Partner with scheduling, orchestration, AI, and software engineers to integrate templates and optimize models for inference and AI applications.

Requirements

• 5–7 years of experience in distributed machine learning with PyTorch or JAX, including multi-node training across 10+ GPUs.
• Expert knowledge of GPU utilization, memory behavior, NCCL collectives, communication bottlenecks, and throughput optimization.
• Hands-on experience debugging convergence issues, profiling bottlenecks, and optimizing distributed training at scale.
• Strong benchmarking methodology, including controlled experiments, noise measurement, variance analysis, and rigorous communication of results.
• Familiarity with TorchTitan, Megatron-LM, or comparable production training frameworks.
• Understanding of FSDP, tensor parallelism, pipeline parallelism, checkpointing, recovery, resource constraints, and cost optimization.

Culture & Benefits

• Full-time employment.
• Inclusive workplace welcoming candidates from diverse backgrounds.
• Opportunity to contribute to sustainable engineering practices in AI infrastructure.
• Reporting to the Head of AI & Applications.