About the company
Sandisk understands how people and businesses consume data and we relentlessly innovate to deliver solutions that enable today’s needs and tomorrow’s next big ideas. With a rich history of groundbreaking innovations in Flash and advanced memory technologies, our solutions have become the beating heart of the digital world we’re living in and that we have the power to shape.
Responsibilities
- Validate neural network operators (matmul, convolution, attention, etc.) for correctness using reference models
- Develop automated tests for operator-level and graph-level validation
- Perform numerical accuracy validation against frameworks like PyTorch or TensorFlow
- Execute and maintain smoke and regression test suites
- Benchmark and validate performance (throughput, latency, scaling behavior)
- Build and maintain performance benchmarking frameworks
- Perform software profiling to analyze execution bottlenecks
- Conduct memory usage analysis and detect memory leaks
- Identify issues across compiler → runtime → kernel execution stack
- Debug failures across compute kernels, runtime APIs, and hardware execution
- Collaborate with compiler, runtime, and hardware teams
Requirements
- 4–6 years in AI/ML systems validation or performance testing
- Strong C++ programming skills (mandatory)
- Strong Object-Oriented Programming (OOP) fundamentals (mandatory)
- Experience with Python for test automation
- Strong understanding of deep learning operators and frameworks (PyTorch preferred)
- Experience with numerical validation and floating-point behavior
- Knowledge of parallel execution, memory hierarchy, and compute kernels
- Preferred: Experience with profiling and debugging tools (perf, VTune, Nsight, etc.)
- Preferred: Experience with ML execution stacks (MLIR/XLA or similar)
- Preferred: Exposure to accelerator validation (GPU/NPU/FPGA)
- Preferred: Experience validating large language model workloads