NVIDIA
Deep Learning Compiler Verification Engineer
$116K–$190K
US, CA, Santa Clara +3 · Full-time · Posted Oct 31, 2026
- Location
- US, CA, Santa Clara
US, TX, Austin
US, OR, Hillsboro
US, WA, Seattle - Experience
- 2+ years of hands-on software engineering experience; relevant experience in compiler development, deep learning systems, or compiler verification.
- Education
- BS, MS or PhD in Computer Science, Computer Engineering, Mathematics, or equivalent experience.
About the job
Develops software and infrastructure to verify the correctness and performance of deep learning compilers across graph transformations, intermediate-representation lowering, and GPU execution.
What you'll do
- Build and maintain systems for reasoning about deep learning compiler correctness and validating optimization semantics and numerical behavior.
- Implement and extend test-generation systems to explore model topologies, precision modes, and hardware targets.
- Improve measurement of functional quality and performance as models, compiler stacks, and hardware evolve.
What they ask for
- Proficiency in Python or C++ and experience with a major deep learning framework, including model execution, graph representation, or runtime behavior.
- Hands-on software engineering experience in compiler development, deep learning systems, or compiler verification.
- Strong problem-solving and debugging skills, including investigating correctness issues and collaborating to resolve edge cases.
Nice to have
- Experience with LLVM, MLIR, TVM, or XLA; compiler passes, IR semantics, or correctness testing.
- Exposure to formal methods or language specification, or experience with quantization, operator fusion, mixed-precision, or graph-level optimization.
Pay
116,000 USD – 189,750 USD base salary
Tools
Python, C++, PyTorch, JAX/XLA, TensorRT, MLIR, LLVM, TVM
AI focus
Builds verification and test-generation software for deep learning compilers and workloads; NVIDIA also states it uses AI tools in recruiting.
Model-written summary · the original posting is authoritative