Lead ML Inference Engineer, Advertising

Roku•San Jose, CA

1d•Hybrid

About The Position

The Advertising Performance group focuses on performance for all participants in the Advertising ecosystem - Advertisers, Publishers, and Roku. The systems and solutions span multiple disciplines and technologies to perform real-time multi-objective optimization across distributed systems at large scale and with low latency. We use Machine Learning, Reinforcement Learning, AI, Control and Optimization Systems, and Auction Dynamics to solve a large set of complex problems. At the core of this is our Machine Learning and Inference Platform that powers the entire landscape. In this role, you will architect, design, and lead the development of a SOTA Inference platform that can handle Advertising-level low latencies, scale, throughput, and availability with optimizations that span across hardware, software, and models. We’re looking for a strong technical leader with deep experience in ML serving, high-performance computing, and industry standard frameworks - someone excited to mentor engineers, innovate at scale, and shape the future of machine learning at Roku.

Requirements

M.S. or above in CS, ECE, or a related field
10+ years of experience in developing and deploying large-scale, distributed systems, with at least 5 years in a leadership or technical lead role
Strong programming skills in high-performance languages
Deep understanding of inference frameworks and ML system deployment
Proven experience optimizing performance for large-scale machine learning systems, including a deep knowledge of SOTA model optimizations, hardware-software co-design, GPU acceleration, and HPC techniques
Excellent communication and collaboration skills
Experience leading teams working on high-throughput, low-latency ML serving systems
Experience collaborating with and leading global, cross-functional teams
Contributions to open-source ML or systems projects

Responsibilities

Lead the design and development of a SOTA Inference platform
Oversee the development of monitoring, observability, and other tooling to ensure system and model performance, reliability, and scalability of online inference services
Identify and resolve system inefficiencies, performance bottlenecks, and reliability issues, ensuring optimized end-end performance
Stay at the forefront of advancements in inference frameworks, ML hardware acceleration, and distributed systems, and incorporate innovations where and when they are impactful