Skip to main content

SambaNova Systems

AI Systems Performance Engineer - New Graduate

San Jose, California, United StatesFrom $165kentryAdded today

About this role

SambaNova is hiring a new graduate AI Systems Performance Engineer to optimize cutting-edge foundation models on their reconfigurable dataflow platform. You'll profile and enhance LLM inference performance across model, compiler, runtime, and hardware layers while collaborating with cross-functional teams.

What you'll do

  • Deploy and bring up foundation models like Llama, Qwen, and DeepSeek on the SambaNova platform
  • Profile model execution to identify performance bottlenecks across the full stack
  • Optimize AI workloads for throughput, latency, and memory efficiency
  • Collaborate with ML, compiler, runtime, and hardware engineers on optimization efforts
  • Develop benchmarks, tools, and performance analysis methodologies for large-scale inference
  • Investigate new model architectures and translate research into efficient production implementations

What they're looking for

  • Python or C++ programming
  • Deep learning frameworks (PyTorch, TensorFlow, or JAX)
  • Computer architecture and systems knowledge
  • Algorithms and data structures
  • Performance profiling and optimization
  • LLM and transformer architecture understanding
  • Model quantization and inference techniques
  • Parallel computing and distributed systems
Apply on the employer's site

Opens the official application on the employer’s site. No login required.

SambaNova Systems

SambaNova Systems builds AI inference and machine learning platforms with specialized compiler infrastructure and runtime optimization. The company is hiring for compiler engineers, software engineers, runtime engineers, and ML solutions engineers to enhance its AI platform capabilities, as well as process and quality engineers for manufacturing excellence.

View all jobs at SambaNova Systems

Likely interview questions

  • Describe a time you optimized a computational system—what metrics did you measure and how did you identify bottlenecks?
  • How would you approach profiling a large language model to find performance bottlenecks across different layers?