Perception Deployment Engineer - Model Deployment & Optimization

Career Tools

About This Role

Zoox is building a ground-up, fully autonomous vehicle fleet along with the supporting ecosystem to deliver urban mobility-as-a-service through robotics and machine learning. The Perception team is currently creating a multi-modality foundation model designed to power the next generation of autonomous vehicle intelligence. This position is ideal for an engineer who wants to integrate large-scale, efficient models directly into vehicle hardware. As a Perception Deployment Engineer focusing on Model Deployment…

Job Description

Responsibilities

• Design and write low-latency, memory-safe, and production-level C++ and CUDA software for real-time perception algorithms on vehicle platforms
• Apply compression techniques such as pruning, quantization aware training, post-training quantization, and mixed-precision frameworks to large-scale models including LLMs, VLMs, and Multi-Modal Sensor Fusion models
• Build and implement compilation and conversion pipelines via TensorRT for edge hardware deployment
• Evaluate compiled edge binaries against PyTorch frameworks through parity checking, latency benchmarking, and accuracy recovery validation
• Write and tune custom machine learning operations and TensorRT Plugins utilizing efficient CUDA kernels to reduce latency and maximize AI accelerator memory bandwidth

Requirements

• Professional programming proficiency in C++ (14, 17, and 20) and Python, with a history of building concurrent, memory-safe, and real-time edge device inference code
• Extensive background in mixed-precision inference frameworks supporting INT8, FP8, BF16, and FP16, alongside model compression methods like PTQ and QAT
• Demonstrated skill in optimizing large-scale models such as LLMs, Multi-Modal Sensor Fusion models, and VLMs or VLAs using KV-cache optimizations and Efficient Attention mechanisms
• Broad experience building model conversion and compilation pipelines using tools like ONNX, TensorRT, and torch.compile while measuring latency benchmarks and parity
• Low-level programming capability for AI accelerators, including the creation and tuning of TensorRT Plugins and custom machine learning operations with CUDA kernels

Qualifications

• Familiarity with state-of-the-art autonomous driving perception algorithms including temporal 3D object detection, BEV, and 3D Occupancy Networks
• Experience processing multi-modal sensor inputs such as Radar, LiDAR, and Vision
• Familiarity with end-to-end autonomous driving paradigms including Foundation models and VLM or VLA models
• Knowledge of edge deployment technologies like TensorRT-LLM

Core Skills

Frequently Asked Questions

Sample Interview Questions

AI-generated questions tailored to this specific role — a preview of the full practice set.