Perception Deployment Engineer - Model Deployment & Optimization
Zoox · Foster City, United States +3 · 9h ago
The Perception team is pioneering the development of a multi-modality foundation model to drive the next generation of autonomous system intelligence.
As a Perception Deployment Engineer, you will focus on bringing highly efficient, production-ready large-scale models to our on-vehicle stack. We are looking for experts with hands-on experience in compressing, accelerating, and deploying complex computer vision or foundation models for power- and thermal-constrained vehicle SOCs. You will optimize the ML models, write custom CUDA kernels, and build highly concurrent inference code to ensure real-time, deterministic execution on edge devices.
In this role, you will:
-
Design and develop production-level, low latency, and memory-safe C++ and CUDA code for real-time perception algorithms on vehicle systems.
-
Optimize large-scale models (Multi-Modal Sensor Fusion models, LLMs, VLMs) using advanced quantization (PTQ, QAT), pruning, mixed-precision inference frameworks.
-
Architect and implement model conversion and compilation pipelines using TensorRT for edge deployment.
-
Perform rigorous parity checking, accuracy recovery, and latency benchmarking between PyTorch frameworks and compiled edge binaries.
-
Develop and optimize custom ML OPs and TensorRT Plugins with efficient CUDA kernels to minimize latency and maximize memory bandwidth on AI accelerators.
Qualifications:
-
Production-level C++ (14/17/20) and Python programming skills, with experience developing concurrent, memory-safe, real-time inference code for edge devices.
-
Deep expertise in model compression technologies (e.g., model quantization such as PTQ and QAT) and mixed-precision inference frameworks (INT8, FP8, BF16/FP16).
-
Proven experience optimizing large-scale models (Multi-Modal Sensor Fusion models, LLMs, VLMs/VLAs) utilizing Efficient Attention mechanisms (e.g., FlashAttention, Linear Attention), KV-cache optimization (e.g., PagedAttention.
-
Extensive experience with model conversion/compilation pipelines (e.g., ONNX, TensorRT, torch.compile) and performing rigorous latency benchmark and model quality parity valuation.
-
Proficiency in low-level programming for AI accelerators, specifically developing and optimizing custom ML OPs and TensorRT Plugins with efficient CUDA kernel implementations.
Bonus Qualifications:
-
Familiarity with SOTA autonomous driving perception algorithms (temporal 3D object detection, BEV, 3D Occupancy Networks) and multi-modal sensor processing (Vision, LiDAR, Radar).
-
Experience with end-to-end autonomous driving paradigms (VLM/VLA models, Foundation models) and edge deployment technologies (e.g., TensorRT-LLM).