On-Device ML Engineer — Mobile AI & Optimization

Unity; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Unity is seeking a Senior Machine Learning Engineer for On-Device & Mobile AI to optimize and deploy cutting-edge models run directly on devices. You’ll own the end-to-end inference stack, from export to kernel-level tuning, targeting mobile and constrained hardware for interactive, real-time experiences.

You will work with WebGPU-based runtimes and native engines, balancing latency, memory footprint, and power while collaborating across research and platform teams to deliver performant,

Qualifications

• 5+ years in software/ML engineering with on‑device or edge inference experience.
• Deployment of transformer- and/or diffusion-based models on mobile/desktop hardware.
• Hands-on with major inference runtimes (ORT Web, CoreML, TFLite, ExecuTorch).
• Low-level performance: GPU/compute APIs and profiling tools.
• Knowledge of quantization, weight sharing, pruning, and distillation.

Responsibilities

• Optimize and deploy inference stacks for constrained hardware.
• Export models, transform graphs, fuse operators, plan memory layouts.
• Tune kernels (WebGPU WGSL, Metal, Vulkan, CUDA) and profile performance.
• Collaborate with research to productionize CV and multi-modal models.
• Benchmark on-device latency, memory, and power within SKUs.

Skills

On-device inference

Transformer models

Diffusion models

GPU/WebGPU

Python

TypeScript/JavaScript

Model optimization

Quantization

Pruning and distillation

Runtime understanding

Collaborative work

Tools

ONNX Runtime Web

CoreML

TFLite

ExecuTorch

WebGPU

Job description

Unity is seeking a Senior Machine Learning Engineer for On-Device & Mobile AI to optimize and deploy cutting-edge models run directly on devices. You’ll own the end-to-end inference stack, from export to kernel-level tuning, targeting mobile and constrained hardware for interactive, real-time experiences.

You will work with WebGPU-based runtimes and native engines, balancing latency, memory footprint, and power while collaborating across research and platform teams to deliver performant,