Senior On-Device ML Engineer – Mobile Inference Expert

Unity; California (MO); On-site; USD 180,000 - 240,000; Full time

Job summary

Unity seeks a Senior ML Engineer for On-Device & Mobile AI to optimize, deploy, and maintain high-performance inference stacks across mobile, desktop, and embedded hardware. You will push efficient multi-modal models—transformers, diffusion, and VLMs—into shipping features with tight latency and memory budgets.

You will work across WebGPU-based runtimes and native options, tuning kernels, and implementing quantization and pruning to meet power and speed targets on diverse SKUs, while

Qualifications

• 5+ years in software/ML engineering with on-device or real-time systems.
• Production deployment of transformer- and/or diffusion-based models on mobile/embedded hardware.
• Hands-on experience with major inference runtimes (ORT Web, CoreML, TFLite, ExecuTorch).
• Low-level performance engineering with GPU/compute APIs and profiling tools.

Responsibilities

• Own the optimization and deployment pipeline for on-device models: export, graph transformation, operator fusion, memory planning, hardware tuning.
• Implement quantization (INT4/INT8/FP16), weight sharing, pruning, and distillation to meet latency and memory budgets.
• Write and tune WebGPU compute shaders and native kernels; profile with browser/platform tools and fix bottlenecks.
• Collaborate with ML researchers to productionize CV and multi-modal architectures for on-device usage.

Skills

On-device inference

Model optimization

WebGPU/WGSL/Metal/Vulkan/CUDA

Profiling & debugging

Python for export pipelines

TypeScript/JavaScript familiarity

Transformer/diffusion deployment

Runtime scheduling

Tools

ONNX Runtime Web

CoreML

TFLite

ExecuTorch

WebGPU

WGSL

Metal

Vulkan

CUDA

Job description

Unity seeks a Senior ML Engineer for On-Device & Mobile AI to optimize, deploy, and maintain high-performance inference stacks across mobile, desktop, and embedded hardware. You will push efficient multi-modal models—transformers, diffusion, and VLMs—into shipping features with tight latency and memory budgets.

You will work across WebGPU-based runtimes and native options, tuning kernels, and implementing quantization and pruning to meet power and speed targets on diverse SKUs, while