ML & Compiler Engineer for AI Accelerators

Amazon Web Services (AWS); Cupertino (CA); On-site; USD 165,000 - 224,000; Full time

Benefits offered by this job

Health insurance

401(k) matching

RSU/stock options

Paid time off

Job summary

Annapurna Labs (U.S.) Inc. is seeking a software engineer to join the AWS Neuron compiler team in Cupertino, CA.

You will build the next generation Neuron compiler, transforming ML models from frameworks like PyTorch, TensorFlow and JAX to run on Inferentia/Trainium in the cloud. You will work with OpenXLA, StableHLO and MLIR, tackle compiler optimization challenges for large language models and vision models, and partner with internal/external customers to deliver high-performance, easy-to-use

Qualifications

• 3+ years of non‑internship professional software development experience.
• 2+ years of design or architecture experience for new and existing systems.
• Experience programming in OO languages such as C++/Java.
• Experience with open-source compiler toolsets like LLVM/MLIR.

Responsibilities

• Design, implement, test, deploy and maintain compiler improvements to optimize Neuron.
• Collaborate with chip architects and ML teams to deploy models on AWS accelerators.
• Work with open-source toolchains (StableHLO/OpenXLA/MLIR) to optimize workloads.
• Contribute features that enhance performance and user experience.

Skills

Software development

OO design

Compiler design

C++/Java

Education

Master's or PhD in CS

Tools

OpenXLA

StableHLO

MLIR

LLVM

PyTorch

JAX

Job description

Annapurna Labs (U.S.) Inc. is seeking a software engineer to join the AWS Neuron compiler team in Cupertino, CA.

You will build the next generation Neuron compiler, transforming ML models from frameworks like PyTorch, TensorFlow and JAX to run on Inferentia/Trainium in the cloud. You will work with OpenXLA, StableHLO and MLIR, tackle compiler optimization challenges for large language models and vision models, and partner with internal/external customers to deliver high-performance, easy-to-use