Software Engineer - ML Ops Framework & Conversion

What You’ll Do

As a Software Engineer on the ML Ops Framework & Conversion team, you will own the pipelines that take models from research to production on edge hardware, including model conversion, compilation, benchmarking, and release.

• Architect and implement model conversion and compilation pipelines using tools such as ONNX, TensorRT, and torch.compile for deployment on edge devices (e.g., NVIDIA Orin).
• Maintain and evolve the model release registry, ensuring traceability and reproducibility across model versions and target platforms.
• Perform rigorous latency benchmarking and model quality parity evaluations to validate that deployed models meet safety-critical performance requirements.
• Compare metrics across platforms to verify accuracy and latency compliance before release.
• Communicate and collaborate with model development teams and broader stakeholders, ensuring your findings translate into reliable, actionable outcomes across the organization.

What You’ll Need to Succeed

• Bachelor's Degree in Computer Science, Electrical Engineering, Robotics, or related field plus 4+ years of relevant experience; or Master's Degree plus 2+ years.
• Extensive experience with model conversion and compilation pipelines (ONNX, TensorRT, torch.compile) and performing rigorous latency benchmarking and quality parity validation.
• Hands-on experience deploying and testing models on edge hardware (e.g., NVIDIA Orin or similar embedded platforms).
• Experience maintaining a model release registry in a production environment.
• Ability to compare and interpret performance metrics across hardware platforms to validate models against strict latency and accuracy requirements.

Bonus Points

• Expertise in model quantization (PTQ, QAT) and mixed-precision inference (INT8, FP8, FP4, BF16/FP16).
• Experience releasing multi-target models across heterogeneous platforms.
• Familiarity with SOTA autonomous driving perception algorithms — temporal 3D object detection, BEV, 3D Occupancy Networks — and multi-modal sensor fusion (vision, LiDAR, radar).
• C++ and/or CUDA kernel development.