Software Engineering Manager, Performance

As a Software Engineering Manager in Performance at Google DeepMind, you will lead and manage a performance engineering team dedicated to enhancing machine learning model performance and efficiency on hardware accelerators. Your role will involve developing team skills, collaborating with various teams, and coaching team members while ensuring effective performance optimization across the software stack.

To qualify as a Software Engineering Manager, Performance at Google DeepMind, candidates should have experience leading engineering teams, strong knowledge of either C++ or Python, and a solid understanding of algorithm and data-structure design. Experience with hardware accelerators, profiling software, leveraging compilers for performance, and knowledge of AI algorithms will give you a competitive edge.

In the Software Engineering Manager, Performance role at Google DeepMind, your work directly contributes to AI advancements by optimizing large language models for training and deployment. Your team's efforts will improve the efficiency and performance of these models, enabling AI researchers to leverage hardware accelerators effectively and push the boundaries of what AI can achieve.

As a Software Engineering Manager, Performance at Google DeepMind, you can expect a vibrant team culture marked by collaboration and creativity. You'll work alongside experienced engineers and research scientists from diverse backgrounds, all united by a shared goal: to enhance the performance of AI technologies while maintaining a commitment to ethics and public benefit.

The base salary range for the Software Engineering Manager, Performance position at Google DeepMind is between $189,000 and $300,000, alongside a bonus, equity, and benefits. The exact salary may vary based on your experience and targeted location, and these details will be discussed further during the hiring process.

As a Software Engineering Manager, it’s essential to balance urgent technical challenges with long-term project goals. I prioritize tasks by assessing their impact on the team and project objectives, ensuring clear communication and delegation based on team strengths.

In my previous roles, I have focused on leveraging profiling tools to identify bottlenecks in large language models, allowing my teams to implement targeted optimizations. This has included custom kernel development and collaborating with AI researchers to refine algorithms for better performance.

I believe in open communication and regular check-ins with AI researchers, establishing a feedback loop that benefits both parties. I emphasize shared goals and encourage team members to actively participate in design discussions where we can intersect engineering with research efforts.

Supporting team growth involves providing mentorship through one-on-one sessions, hands-on workshops, and setting clear opportunities for professional development. I also encourage team members to share their own knowledge through collaborative projects.

I approach performance bottlenecks systematically by analyzing profiling data, identifying root causes, and collaboratively designing solutions with my team, ensuring that optimizations align with our broader performance objectives.

I have extensive experience with CUDA for programming GPUs, which has facilitated the design of optimized kernels tailored to our hardware for enhanced performance in ML applications.

My coaching approach involves thorough onboarding, setting clear expectations, and pairing new members with experienced engineers for hands-on guidance. I focus on creating an inclusive environment where questions are encouraged and ongoing development is supported.

I evaluate new technologies based on their scalability, ease of integration with our existing systems, and the potential for significant performance gains. Additionally, I consider the community support and documentation available for any new tools.

I stay current with advancements through continuous learning, attending conferences, enrolling in advanced courses, and actively following key publications and industry leaders in ML and hardware acceleration.

In my previous role, we encountered delays in model deployment due to performance optimization issues. I organized a collaborative sprint involving cross-functional teams to address the issues aggressively, resulting in timely delivery without compromising quality.