Edge AI Perception Deployment Engineer for Real-Time Models

Lever, Inc.; Boston (MA); On-site; USD 199,000 - 270,000; Full time

Benefits offered by this job

Health insurance

Zoox Stock Appreciation Rights

Amazon RSUs

Paid time off

Disability insurance

Life insurance

Sign-on bonus (may be offered)

Job summary

Zoox is building the first fully autonomous vehicle fleet with a strong focus on on-vehicle, edge deployment. We are seeking a Perception Deployment Engineer to bring efficient large-scale models to production on power- and thermal-constrained SOCs.

You will craft low-latency C++/CUDA code, optimize Multi-Modal Sensor Fusion models and LLMs, and implement TensorRT-based pipelines for real-time perception on edge devices. Strong emphasis on parity, accuracy recovery, and memory efficiency.

Qualifications

• Production-level C++ (14/17/20) and Python programming for real-time edge inference.
• Deep expertise in model compression (PTQ, QAT) and mixed-precision inference (INT8, FP8, BF16/FP16).
• Experience optimizing large-scale models (Multi-Modal Sensor Fusion, LLMs, VLMs) with Efficient Attention and KV-cache optimization.
• Model conversion/compilation pipelines (ONNX, TensorRT, torch.compile) and latency parity evaluation.
• Low-level programming for AI accelerators; develop and optimize custom ML OPs and TensorRT plugins with CUDA kernels.

Responsibilities

• Design and develop production-level C++/CUDA code for real-time perception on vehicle systems.
• Optimize large-scale models using quantization and mixed-precision techniques.
• Build and optimize TensorRT-based edge deployment pipelines.
• Perform parity checks and latency benchmarking between PyTorch and compiled edge binaries.
• Develop custom ML OPs and TensorRT plugins with efficient CUDA kernels.

Skills

C++

Python

Concurrent programming

Memory-safe real-time inference

Model compression (PTQ/QAT)

Tools

ONNX

TensorRT

torch.compile

CUDA

Job description

Zoox is building the first fully autonomous vehicle fleet with a strong focus on on-vehicle, edge deployment. We are seeking a Perception Deployment Engineer to bring efficient large-scale models to production on power- and thermal-constrained SOCs.

You will craft low-latency C++/CUDA code, optimize Multi-Modal Sensor Fusion models and LLMs, and implement TensorRT-based pipelines for real-time perception on edge devices. Strong emphasis on parity, accuracy recovery, and memory efficiency.