Staff Software Engineer, AI Platform

At LinkedIn, our approach to flexible work is centered on trust and optimized for culture, connection, clarity, and the evolving needs of our business. The work location of this role is hybrid, meaning it will be performed both from home and from a LinkedIn office on select days, as determined by the business needs of the team. 

Job Description

This role can be based in Mountain View, CA, San Francisco, CA, or Bellevue, WA.

Join us to push the boundaries of scaling large models together. The team is responsible for scaling LinkedIn's AI model training, feature engineering and serving with hundreds of billions of parameters models and large scale feature engineering infra for all AI use cases from recommendation models, large language models, to computer vision models. We optimize performance across algorithms, AI frameworks, data infra, compute software, and hardware to harness the power of our GPU fleet with thousands of latest GPU cards. The team also works closely with the open source community and has many open source committers (TensorFlow, Horovod, Ray, vLLM, Hugginface, DeepSpeed etc.) in the team. Additionally, this team focussed on technologies like LLMs, GNNs, Incremental Learning, Online Learning and Serving performance optimizations across billions of user queries.

Model Training Infrastructure: As an engineer on the AI Training Infra team, you will play a crucial role in building the next-gen training infrastructure to power AI use cases. You will design and implement high performance data I/O, work with open source teams to identify and resolve issues in popular libraries like Huggingface, Horovod and PyTorch, enable distributed training over 100s of billions of parameter models, debug and optimize deep learning training, and provide advanced support for internal AI teams in areas like model parallelism, tensor parallelism, Zero++ etc. Finally, you will assist in and guide the development of containerized pipeline orchestration infrastructure, including developing and distributing stable base container images, providing advanced profiling and observability, and updating internally maintained versions of deep learning frameworks and their companion libraries like Tensorflow, PyTorch, DeepSpeed, GNNs, Flash Attention. PyTorch Lightning and more.

Feature Engineering: this team shapes the future of AI with the state-of-the-art Feature Platform, which empowers AI Users to effortlessly create, compute, store, consume, monitor, and govern features within online, offline, and nearline environments, optimizing the process for model training and serving. As an engineer in the team, you will explore and innovate within the online, offline, and nearline spaces at scale (millions of QPS, multi terabytes of data, etc), developing and refining the infrastructure necessary to transform raw data into valuable feature insights. Utilizing leading open-source technologies like Spark, Beam, and Flink and more, you will play a crucial role in processing and structuring feature data, ensuring its most optimal storage in the Feature Store, and serving feature data with high performance.

Model Serving Infrastructure: this team builds low latency high performance applications serving very large & complex models across LLM and Personalization models. As an engineer, you will build compute efficient infra on top of native cloud, enable GPU based inference for a large variety of use cases, cuda level optimizations for high performance, enable on-device and online training. Challenges include scale (10s of thousands of QPS, multiple terabytes of data, billions of model parameters), agility (experiment with hundreds of new ML models per quarter using thousands of features), and enabling GPU inference at scale.

ML Ops: The MLOps and Experimentation team is responsible for the infrastructure that runs MLOps and experimentation systems across LinkedIn. From Ramping to Observability, this org powers the AI products that define LinkedIn. This team, inside MLOps, is responsible for AI Metadata, Observability, Orchestration, Ramping and Experimentation for all models; building tools that enable our product and infrastructure engineers to optimize their models and deliver the best performance possible.

As a Staff Software Engineer, you will have first-hand opportunities to advance one of the most scalable AI platforms in the world. At the same time, you will work together with our talented teams of researchers and engineers to build your career and your personal brand in the AI industry.

Responsibilities

• Owning the technical strategy for broad or complex requirements with insightful and forward-looking approaches that go beyond the direct team and solve large open-ended problems.
• Designing, implementing, and optimizing the performance of large-scale distributed serving or training for personalized recommendation as well as large language models.
• Improving the observability and understandability of various systems with a focus on improving developer productivity and system sustenance.
• Mentoring other engineers, defining our challenging technical culture, and helping to build a fast-growing team.
• Working closely with the open-source community to participate and influence cutting edge open-source projects (e.g., vLLMs, PyTorch, GNNs, DeepSpeed, Huggingface, etc.).
• Functioning as the tech-lead for several concurrent key initiatives AI Infrastructure and defining the future of AI Platforms.

Basic Qualifications

• Bachelor's Degree in Computer Science or related technical discipline, or equivalent practical experience
• 4+ years of experience in the industry with leading/ building deep learning systems.
• 4+ years of experience with Java, C++, Python, Go, Rust, C# and/or Functional languages such as Scala or other relevant coding languages
• Hands-on experience developing distributed systems or other large-scale systems.

Preferred Qualifications

• BS and 8+ years of relevant work experienceMS and 7+ years of relevant work experience, or PhD and 4+ years of relevant work experience
• Previous experience working with geographically distributed co-workers.
• Outstanding interpersonal communication skills (including listening, speaking, and writing) and ability to work well in a diverse, team-focused environment with other SRE/SWE Engineers, Project Managers, etc.
• Experience building ML applications, LLM serving, GPU serving.
• Experience with search systems or similar large-scale distributed systems
• Expertise in machine learning infrastructure, including technologies like MLFlow, Kubeflow and large scale distributed systems
• Experience with distributed data processing engines like Flink, Beam, Spark etc., feature engineering,
• Co-author or maintainer of any open-source projects
• Familiarity with containers and container orchestration systems
• Expertise in deep learning frameworks and tensor libraries like PyTorch, Tensorflow, JAX/FLAX

Suggested Skills

• ML Algorithm Development
• Machine Learning and Deep Learning
• Information retrieval / recommendation systems / distributed serving / Big Data is a plus.
• Communication
• Stakeholder Management

You will Benefit from our Culture

We strongly believe in the well-being of our employees and their families. That is why we offer generous health and wellness programs and time away for employees of all levels.

LinkedIn is committed to fair and equitable compensation practices. The pay range for this role is $170,000 - $277,000. Actual compensation packages are based on several factors that are unique to each candidate, including but not limited to skill set, depth of experience, certifications, and specific work location. This may be different in other locations due to differences in the cost of labor.

The total compensation package for this position may also include annual performance bonus, stock, benefits and/or other applicable incentive compensation plans. For additional information, visit: https://careers.linkedin.com/benefits.

Team Name: DragonKnight (Pytorch & LLM Training) Team

Team Description: We build LLM training and post-training frameworks to support scalable and efficient LLM model training and optimizations. We also optimize LLM training by developing customized and efficient Triton kernels to improve training throughput and cut memory footprint.

Equal Opportunity Statement 

We seek candidates with a wide range of perspectives and backgrounds and we are proud to be an equal opportunity employer. LinkedIn considers qualified applicants without regard to race, color, religion, creed, gender, national origin, age, disability, veteran status, marital status, pregnancy, sex, gender expression or identity, sexual orientation, citizenship, or any other legally protected class.

LinkedIn is committed to offering an inclusive and accessible experience for all job seekers, including individuals with disabilities. Our goal is to foster an inclusive and accessible workplace where everyone has the opportunity to be successful.

If you need a reasonable accommodation to search for a job opening, apply for a position, or participate in the interview process, connect with us at [email protected] and describe the specific accommodation requested for a disability-related limitation.

Reasonable accommodations are modifications or adjustments to the application or hiring process that would enable you to fully participate in that process. Examples of reasonable accommodations include but are not limited to:

• Documents in alternate formats or read aloud to you
• Having interviews in an accessible location
• Being accompanied by a service dog
• Having a sign language interpreter present for the interview

A request for an accommodation will be responded to within three business days. However, non-disability related requests, such as following up on an application, will not receive a response.

LinkedIn will not discharge or in any other manner discriminate against employees or applicants because they have inquired about, discussed, or disclosed their own pay or the pay of another employee or applicant. However, employees who have access to the compensation information of other employees or applicants as a part of their essential job functions cannot disclose the pay of other employees or applicants to individuals who do not otherwise have access to compensation information, unless the disclosure is (a) in response to a formal complaint or charge, (b) in furtherance of an investigation, proceeding, hearing, or action, including an investigation conducted by LinkedIn, or (c) consistent with LinkedIn's legal duty to furnish information.

San Francisco Fair Chance Ordinance ​

Pursuant to the San Francisco Fair Chance Ordinance, LinkedIn will consider for employment qualified applicants with arrest and conviction records.

Pay Transparency Policy Statement ​

As a federal contractor, LinkedIn follows the Pay Transparency and non-discrimination provisions described at this link: https://lnkd.in/paytransparency.

Global Data Privacy Notice for Job Candidates ​

Please follow this link to access the document that provides transparency around the way in which LinkedIn handles personal data of employees and job applicants: https://legal.linkedin.com/candidate-portal.