AI Engineer (PyTorch)

Location: Must be based in or be able to relocate to Almaty, Kazakhstan (On-site)

Company

hirify.global is a rapidly scaling generative AI company powering creative tools for Fortune 500 brands with a focus on AI-native video experiences.

What you will do

• Profile end-to-end training runs to identify bottlenecks across compute, memory, and storage.
• Optimize distributed training strategies including data, tensor, pipeline, and expert parallelism.
• Develop and integrate optimized CUDA and Triton kernels.
• Improve collective communication through topology-aware placement.
• Diagnose distributed hangs and performance regressions.
• Enhance fault tolerance for long-running training jobs.

Requirements

• Strong experience running and optimizing multi-GPU or multi-node training.
• Proficiency with PyTorch Distributed or equivalent training frameworks.
• Deep understanding of GPU architecture, memory hierarchy, and Tensor Cores.
• Experience with distributed parallelism technologies like FSDP, DeepSpeed, or Megatron-LM.
• Ability to debug complex performance and reliability issues across the training stack.
• Must be willing to work on-site in Almaty, Kazakhstan.

Nice to have

• CUDA or Triton kernel development experience.
• Experience with NCCL, MPI, RDMA, InfiniBand, or NVLink.
• Knowledge of Mixture-of-Experts, multimodal, or reinforcement learning models.
• Experience with PyTorch internals, torch.compile, or ML compilers.

Culture & Benefits

• Competitive base salary in USD.
• Equity participation through a stock option program.
• Relocation support provided for candidates moving to Almaty.
• Collaborative, fast-paced environment working directly with experienced leadership.
• Office benefits including meals and transportation.