Senior AI Inference Engineer - Model Optimization & Deployment

The Perception team is pioneering the development of a multi-modality foundation model to drive the next generation of autonomous system intelligence. As a Model Optimization & 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 models (LLMs, VLMs, or FMs) 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: Optimize large-scale models (Multi-Modal Sensor Fusion models, LLMs, VLMs) using advanced quantization (PTQ, QAT), pruning, mixed-precision inference frameworks, and parameter-efficient fine-tuning (LoRA, QLoRA). 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. Write production-level, low latency, and memory-safe C++ and CUDA code for real-time inference on vehicle systems. Qualifications: Deep expertise in model quantization (PTQ, QAT) and mixed-precision inference frameworks (INT8, FP8, FP4, 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) and Speculative Decoding. 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. Production-level C++ (14/17/20) and Python programming skills, with experience developing concurrent, memory-safe, real-time inference code for edge devices. 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 distributed training pipelines and model/tensor parallelism (PyTorch Distributed, Ray, DeepSpeed, Megatron-LM) and runtime efficiency optimization for GPU clusters. Experience with end-to-end autonomous driving paradigms (VLM/VLA models, Foundation models) and edge deployment technologies (e.g., TensorRT-LLM). About Zoox Zoox is developing the first ground-up, fully autonomous vehicle fleet and the supporting ecosystem required to bring this technology to market. Sitting at the intersection of robotics, machine learning, and design, Zoox aims to provide the next generation of mobility-as-a-service in urban environments. We’re looking for top talent that shares our passion and wants to be part of a fast-moving and highly execution-oriented team. Follow us on LinkedIn Accommodations If you need an accommodation to participate in the application or interview process please reach out to accommodations@zoox.com or your assigned recruiter. A Final Note: You do not need to match every listed expectation to apply for this position. Here at Zoox, we know that diverse perspectives foster the innovation we need to be successful, and we are committed to building a team that encompasses a variety of backgrounds, experiences, and skills.