AI Systems, Training

Responsibilities

• Build and maintain highly optimized, model-specific training stacks specifically tuned for state-of-the-art generative vision, language, and world models.
• Design and scale multi-node distributed training systems, implementing elastic sharding and robust data streaming pipelines for fast, large-scale iteration.
• Implement robust model checkpointing and recovery mechanisms.
• Develop and optimize kernels using low-level programming models like CUDA and Triton.
• Design rigorous benchmarking suites to track Model Flops Utilization (MFU), memory bandwidth, and convergence stability.
• Act as a translator, discussing algorithmic trade-offs with theorists and converting model requirements into concrete specifications for infrastructure and hardware engineering teams.

Qualifications

• An MS/PhD or equivalent research/project experience in a quantitative field such as AI/Machine Learning, Computer Science, Physics, Electrical Engineering, or Applied Math.
• Veteran of the modern ML software stack.
• Demonstrated ability to map state-of-the-art AI model architectures (e.g., transformers, Mixture of Experts, diffusion models) to system performance implications.
• Deep expertise in how models are partitioned across a cluster, with a mastery of communication primitives and parallelism strategies.
• Proven track record of implementing, debugging, and maintaining production-grade training frameworks—such as Megatron-LM, DeepSpeed, Ray, PyTorch Lightning—turning raw compute into a reliable model-building factory.