Foundation Models Scientist: AI Systems & GPU Optimization

Amazon Web Services (AWS); Santa Clara (CA); On-site; USD 172,000 - 222,000; Full time

Job summary

Amazon Web Services, Inc. is seeking an AI/ML Engineer to build efficient, stable foundation models for long-horizon agentic workloads. You will design, train, and evaluate models, while optimizing for latency, throughput, and memory usage across GPU and accelerator hardware.

The role combines model development with systems and hardware optimization, requiring deep expertise in ML/DL, transformers, and parallel training. Collaboration across teams and a strong publishing record are encouraged.

Qualifications

• Strong background in ML, DL, transformers, and foundation models.
• Experience with PyTorch, JAX, or similar frameworks.
• Experience training, fine-tuning, or serving large-scale models.
• Knowledge of GPU architecture, CUDA, Triton, custom kernels, or hardware-aware optimization.
• Familiarity with distributed training and model parallelism.
• Experience with AWS Trainium, Inferentia, Neuron SDK, TPUs, or other custom accelerators is preferred.
• PhD in ML, CS, EE, or related field is preferred.

Responsibilities

• Design, train, fine‑tune, and evaluate custom foundation models.
• Improve model efficiency, stability, latency, throughput, and memory usage.
• Build models that support planning, tool use, memory, self‑correction, and multi‑step task execution.
• Optimize training and inference using CUDA, Triton, custom GPU kernels, and tensor operations.
• Profile and resolve compute, memory, communication, and distributed‑training bottlenecks.
• Work with GPUs and specialized AI accelerators, including AWS Trainium and other custom tensor hardware.
• Develop evaluations for model reliability and long‑horizon agent performance.

Skills

Machine learning

Deep learning

Transformers

Foundation models

PyTorch

JAX

CUDA

Distributed training

Python

C++

GPU architecture

Education

PhD in ML/CS/EE

MS in CS/CE/ML

Tools

Triton

Custom kernels

Tensor operations

AWS Trainium/Inferentia

Neuron SDK

TPUs

GPU hardware optimization

Job description

Amazon Web Services, Inc. is seeking an AI/ML Engineer to build efficient, stable foundation models for long-horizon agentic workloads. You will design, train, and evaluate models, while optimizing for latency, throughput, and memory usage across GPU and accelerator hardware.

The role combines model development with systems and hardware optimization, requiring deep expertise in ML/DL, transformers, and parallel training. Collaboration across teams and a strong publishing record are encouraged.