Deep Learning Compiler Verification Engineer

NVIDIA•Santa Clara, CA

About The Position

NVIDIA is building next-generation compiler technologies to accelerate deep learning workloads and is seeking an engineer to develop compiler verification software and related infrastructure. The role involves solving correctness challenges in GPU computing and systems software, working with compiler and architecture teams to evaluate optimizations and their effects on program semantics and numerical behavior.

Requirements

BS, MS, or PhD in Computer Science, Computer Engineering, Mathematics, or equivalent experience.
At least 2 years of hands-on software engineering experience, including experience in compiler development, deep learning systems, or compiler verification.
Proficiency in Python or C++.
Experience with a major deep learning framework such as PyTorch, JAX/XLA, TensorRT, or a similar framework, involving model execution, graph representation, or runtime behavior.
Strong problem-solving and debugging skills, including the ability to reason about model behavior and compiler-generated code, investigate correctness issues, and collaborate to resolve edge-case failures.

Nice To Haves

Experience with compiler technologies such as LLVM, MLIR, TVM, or XLA, and familiarity with compiler passes, IR semantics, or correctness testing.
Exposure to formal methods or language specification, such as type systems, program semantics, or proof-based verification.
Experience with deep learning model internals such as quantization, operator fusion, mixed precision, or graph-level optimization.

Responsibilities

Develop and maintain systems components for reasoning about correctness in deep learning compilers across graph transformations, IR lowering, and GPU execution.
Work with deep learning compiler and architecture teams to analyze and validate optimizations, including MLIR graph rewrites, fusion passes, and mixed-precision transformations, ensuring they preserve semantics and numerical behavior.
Implement and extend test-generation systems that use deep learning solutions and analysis methods to drive in-depth testing across model topologies, precision modes, and hardware targets.
Contribute to improving how functional quality and performance are measured as models, compiler stacks, and hardware evolve.