About The Position

Microsoft Foundry helps developers and organizations build, deploy, and operate AI solutions at scale. The engineering team builds infrastructure to make model training and inference faster, more reliable, and more efficient for customers. This role joins the Foundry training team to improve pre-training and post-training systems. The engineer will own a training component and help optimize agents using interaction trajectories, tool-use data, and feedback. Microsoft is hiring at both Software Engineer II and Senior Software Engineer levels. The position involves profiling and optimizing training-system performance, shipping tested changes with reproducible benchmarks, and reducing training step time and memory use while preserving restart correctness, model quality, and numerical correctness. It also involves building secure, tenant-isolated systems with automated testing, observability, reliable recovery, and safe releases, as well as providing operational support and improving services using production feedback. The role aims to improve throughput, latency, and cost, pursue near-linear throughput scaling where workloads permit across agreed GPU and node configurations, and validate convergence and held-out model and agent quality. Quality validation includes multi-turn task completion, tool-use correctness, and reward-exploitation checks for reinforcement learning.

Requirements

  • Bachelor's degree in Computer Science or a related technical field and at least 2 years of technical engineering experience coding in languages including C, C++, C#, Java, JavaScript, or Python, or equivalent experience.
  • Ability to meet Microsoft, customer, and/or government security screening requirements, including passing the Microsoft Cloud Background Check upon hire or transfer and every two years thereafter.

Nice To Haves

  • At least 5 years of technical engineering experience coding in languages including C, C++, C#, Java, JavaScript, or Python, or equivalent experience.
  • Python and hands-on training implementation using PyTorch, JAX, or comparable frameworks.
  • Proven software engineering fundamentals, debugging skills, and clear cross-team communication.
  • Experience diagnosing performance bottlenecks in training workloads.
  • Experience with Kubernetes, Ray, and training toolchains such as PyTorch, JAX, Slime, verl, or TRL.

Responsibilities

  • Use profiling to improve data loading, sequence packing, activation memory, or checkpoint I/O.
  • Ship tested optimizations and reproducible benchmarks.
  • Reduce step time and memory use while preserving restart correctness and training quality.
  • Build secure, tenant-isolated systems with automated testing, observability, reliable recovery, and safe releases.
  • Participate in operational support and improve services using production feedback.
  • Preserve model quality and numerical correctness.
  • Use reproducible benchmarks to improve throughput, latency, and cost.
  • Pursue near-linear throughput scaling where workloads permit across agreed GPU and node configurations.
  • Validate convergence and held-out model and agent quality, including multi-turn task completion, tool-use correctness, and reward-exploitation checks for RL.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service