Description

You will design, build, and optimize the full stack required for large-scale model training, fine-tuning, and inference to maximize Model Flop Utilization (MFU) on compute clusters.

Responsibilities

• Design and optimize dataloading, distributed training, and model deployment pipelines.
• Translate research algorithms into production-grade, high-performance code and scalable infrastructure.
• Profile training workflows to identify and resolve bottlenecks from input pipelines to inference.
• Implement advancements from research publications and open-source contributions into enterprise systems.
• Use MLOps tools to establish best practices for model development, validation, deployment, and monitoring.

Required Skills

• 5+ years of experience building and optimizing large-scale machine learning infrastructure.
• Expertise in Python and deep learning frameworks including PyTorch and Jax.
• Hands-on experience with distributed training and data pipelines in multi-node GPU environments.
• Proven ability to scale models trained on large datasets across distributed computing systems.
• Experience with TensorFlow and general Machine Learning/Artificial Intelligence principles.
• Proficiency with MLOps frameworks and performance profiling tools for lifecycle management.
• Ability to collaborate with research scientists to integrate models into production.
• Degree in B.E., B.Tech, M.E., M.Tech, BCA, MCA, or equivalent.

Preferred Skills

• Experience writing and optimizing CUDA, Triton, or CUTLASS kernels.
• Track record of open-source contributions in AI infrastructure or data engineering.

Key Skills

Education

Any Graduate