ML Infra Engineer, Modeling

The Team

The ML Infrastructure team supports and accelerates PI’s core modeling efforts by building the systems that make large-scale training reliable, reproducible, and fast. The team works closely with research, data, and platform engineers to ensure models can scale from prototype to production-grade training runs.

In This Role You Will

• Own training/inference infrastructure: Design, implement, and maintain systems for large-scale model training, including scheduling, job management, checkpointing, and metrics/logging.
• Scale distributed training: Work with researchers to scale JAX-based training across TPU and GPU clusters with minimal friction.
• Optimize performance: Profile and improve memory usage, device utilization, throughput, and distributed synchronization.
• Enable rapid iteration: Build abstractions for launching, monitoring, debugging, and reproducing experiments.
• Partner with researchers: Translate research needs into infra capabilities and guide best practices for training at scale.
• Contribute to core training code: Evolve JAX model and training code to support new architectures, modalities, and evaluation metrics.

What We Hope You’ll Bring

• Strong software engineering fundamentals and experience building ML training infrastructure or internal platforms.
• Hands-on large-scale training experience in JAX (preferred), PyTorch.
• Familiarity with distributed training, multi-host setups, data loaders, and evaluation pipelines.
• Experience managing training workloads on cloud platforms (e.g., SLURM, Kubernetes, GCP TPU/GKE, AWS).
• Ability to debug and optimize performance bottlenecks across the training stack.
• Strong cross-functional communication and ownership mindset.

Bonus Points If You Have

• Deep ML systems background (e.g., training compilers, runtime optimization, custom kernels).
• Experience operating close to hardware (GPU/TPU performance tuning).
• Background in robotics, multimodal models, or large-scale foundation models.
• Experience designing abstractions that balance researcher flexibility with system reliability.