Senior AI/ML Engineer

What you'll do

• Validate optimized model implementations by defining equivalence criteria and tests that expose numerical violations.
• Connect numerical differences to open- and closed-loop driving outcomes, and identify the causes of behavioral disparity.
• Develop sensitivity, robustness, and training-diagnostics tools that run efficiently across repeated and distributed training jobs.

What they ask for

• Working command of numerical analysis and matrix theory, including Jacobian/Hessian estimation, spectral properties, conditioning, and floating-point error analysis.
• Strong understanding of neural-network training, including loss landscapes, gradient and error propagation, and optimizer dynamics.
• Adversarial approach to numerical edge cases such as denormals, extreme dynamic range, catastrophic cancellation, and accumulation-order effects.
• High proficiency in PyTorch and Python, with experience building mathematically defensible analytical tools fast enough for production training and evaluation loops.

Nice to have

• Experience debugging large-scale training runs or distributed training beyond DDP, including FSDP, Megatron-LM, DeepSpeed, or 3D parallelism.
• Experience with quantization, compiler toolchains, inference-time numerical parity, AV/ADAS evaluation, or robustness and OOD analysis.

Pay

$170,600 - $261,300 USD base salary

Tools

PyTorch, Python, DDP, FSDP, Megatron-LM, DeepSpeed, WSD

AI focus

Analyzes and improves the numerical reliability of compressed, quantized, compiled, and precision-reduced autonomous-driving models; links numerical changes to behavioral outcomes and diagnoses model-training instabilities.