About
- Engineer Manager of Training team in Uber AI platform (Michelangelo).
- Expert in…
- Expert in…
Articles by Peng
Activity
3K followers
-
Peng Zhang posted this字节跳动 AML Engine 编排调度组最近在 Seattle 或 SJC 招聘如下岗位。 主要从事 TikTok 搜索，广告，推荐模型 的 GPU infra 编排调度 （scheduling and orchestration)，管理百万卡量级GPU集群，训练和推理都招。 希望有 k8s operators/controllers， 资源池管理，新GPU硬件上线，动态扩缩容，A/B test 流量切换，QPS压力测试，训练workflow，模型封装和deployment， 跨region容灾 等等。 要求有 ML platform，GPU scheduling， K8s 集群管理，Compute resource orchestration， multi-host GPU training/inferernce 等相关工作/实习经验。 校招/实习也看中有 open source开发经验的同学, 比如 Ray，Airflow，Kubeflow， Flyte， Kueue，Temporal, Skypilot, SGLang, etc。 Prefer 可以中英双语交流 （常与中国区会议）， 对OCI/GCP 有经验的同学。 - Sr/Staff SDE 或者 工作经验特别match的SDE2，请点下面link申请。也欢迎直接联系我了解情况。 ---- 西雅图： https://lnkd.in/gSNbQpQj ---- SJC：https://lnkd.in/gc-HrbFj - SWE New Grad 直接点以下链接申请 ---- 西雅图 BS/MS：https://lnkd.in/g-vWps7J ---- SJC BS/MS：https://lnkd.in/gEgDW_A7 ---- 西雅图PhD：https://lnkd.in/gjfUqP-r ---- SJC PhD：https://lnkd.in/gTftMgAf - SWE intern 直接点以下链接申请 ---- BS/MS intern in SEA/SJC: https://lnkd.in/g3WfdpzF ---- PHD intern in SEA/SJC: https://lnkd.in/gKpHCRFg ---- 非常match但没有CPT的同学，可帮���联系国内实习机会，之后转美国全职岗位
-
Peng Zhang reposted thisPeng Zhang reposted thisMost infrastructure providers buy GPUs and rent them out. We build every layer – the power, the data centers, the compute. Today, Nscale has agreed to acquire Anyscale, the platform behind Ray, built for scaling AI workloads across thousands of GPUs. We built the infrastructure AI runs on. Anyscale built the software developers actually live in. Together, we go from capacity supplier to the full-stack home for AI development – power to production. Anyscale will keep operating under its own brand, serving its customers as it does today. And we're joining the PyTorch Foundation, standing behind Ray's open governance and the community that built it. What excites me most: Anyscalers joining us. That's the caliber of talent shaping what comes next. I get to work with more talented people!! Most companies optimize one layer. We're building all of them, on purpose. More here: https://lnkd.in/gUbfnpa3
-
Peng Zhang shared thisMichelangelo’s submission to PyTorch Conf has been accepted. We’re excited to have the team present the impactful work we’ve accomplished on top of PyTorch, Ray and DeepSpeed open sources. Topic: “PyTorch-Native Feature Transformation and Training Framework for Uber Eats Recommendation” PS: Will co-present this work as ex-Uber.
-
Peng Zhang shared thisWe are hiring Software Engineers in multiple levels in Uber AI platform ( Michelangelo) team in Seattle, WA or Sunnyvale, CA. Please apply directly if interested. https://lnkd.in/g2MsBfSM https://lnkd.in/gVJmd7QV https://lnkd.in/gQGpKw2J https://lnkd.in/geZ-NmXw
-
Peng Zhang shared thisGreat to see Hotels in Uber besides foods and drinks. I really like the direction! Hope some day in future we can use Ube to book tickets like parks, events, museums, games, concerts, etc. That will make it a one-stop shop for a good travel.Peng Zhang shared thisHotels on Uber has arrived! Today, we announced the launch of Hotels on Uber at our premiere consumer event GO–GET. Our goal is to be ever more helpful in your travel journey. Most of us already rely on Uber to get to the airport, get to the hotel, go around the city, order food and groceries. With Hotels, now you can complete that journey all in one app. So what’s in it for you? We are bringing the same experience - easy, clear, uncomplicated, that you are accustomed to while booking a ride - to booking a hotel. And there is huge value that I think you’ll love. 20% off on most hotels and Uber One members get an additional 10% back in Uber One Credits. On your next vacation, try Uber Hotels, it will cover your rides to the airport and those meals that you order on your trip. We’re rolling this out in the US over the next few days – let me know what you think. I am very proud of the team behind this, who worked extra hard to bring this value to all of you!
-
Peng Zhang shared thisExcited to share that I recently presented at Ray Day Seattle 2026 on scaling large-scale machine learning training 🚀 In my talk, “Scale 10B+ Model Training with Ray,” I shared how we’re evolving Uber’s AI platform to support the next generation of deep learning workloads — scaling from millions to 10B+ parameter models. Some key takeaways: - Building a distributed training stack with Ray, PyTorch, and DeepSpeed to efficiently scale across heterogeneous clusters - Evolving from traditional pipelines to Ray Data to improve GPU utilization and eliminate data loading bottlenecks - Leveraging model parallelism (ZeRO), mixed precision, and Flash Attention to push model size and training efficiency - Achieving up to 20% improvement in GPU utilization and ~50% reduction in training time in large-scale pipelines - It’s exciting to see how the Ray ecosystem is maturing — enabling a more unified approach where data processing and training pipelines converge into a single system. Big thanks to the Uber AI Platform (Michelangelo) team for driving this work forward, and to Anyscale for hosting this fantastic event Robert Nishihara Brent Bain. If you’re working on large-scale ML systems or distributed training, I’d love to connect and exchange ideas. Also enjoyed connecting with and learning from industry peers in Seattle — Chi Wang Mickey Liu Haocheng B. Qia Wang Shujun Bian https://lnkd.in/gHqSbcGw https://lnkd.in/gCKGd5SJ https://lnkd.in/g77ZF4xs #MachineLearning #DistributedSystems #Ray #DeepLearning #MLOps #AIInfrastructure
-
Peng Zhang reposted thisPeng Zhang reposted thisExplore how we optimized Petastorm to cut deep learning training time by 6× and eliminated hidden randomness at Uber. Read more: https://lnkd.in/g4sXsKMM #Uber #UberEngineering #MachineLearningAccelerating Deep Learning: How Uber Optimized Petastorm for High-Throughput and Reproducible GPU TrainingAccelerating Deep Learning: How Uber Optimized Petastorm for High-Throughput and Reproducible GPU Training
-
Peng Zhang reposted thisPeng Zhang reposted thisLearn how Uber Eats scaled its homefeed discovery engine by transitioning to transformer-based generative recommenders and a modernized ML infrastructure to deliver real-time, personalized recommendations to millions. Read More: https://lnkd.in/g_pJwt2C #Uber #UberEngineering #MachineLearningNext-Gen Restaurant Recommendation with Generative Modeling and Real-Time FeaturesNext-Gen Restaurant Recommendation with Generative Modeling and Real-Time Features
-
Peng Zhang reposted thisPeng Zhang reposted thisSome of the great work our AI infra and customer teams (led by Himaanshu Gupta , Peng Chen and many other folks). Congrats to everyone! https://lnkd.in/gmn5mJunNext-Gen Restaurant Recommendation with Generative Modeling and Real-Time FeaturesNext-Gen Restaurant Recommendation with Generative Modeling and Real-Time Features
Peng Zhang replied to a comment
1w
Shen Li 是的，这个岗位是GPU infra的编排调度方面。
-
Peng Zhang liked thisPeng Zhang liked this
-
Peng Zhang liked thisPeng Zhang liked thisI’m excited to be speaking at AI Agenda Live in San Francisco on September 23, joining leaders across the AI ecosystem to explore what’s next for the technology—and what’s actually happening on the ground. If you’d like to attend, register here: https://lnkd.in/dEj9P_9F
-
Peng Zhang liked thisPeng Zhang liked thisHarvey is hiring a Staff Product Manager, Infrastructure (San Francisco). This is the person who'll own the roadmap for the systems powering every interaction on our legal AI platform — multi-region deployments, observability, and reliability/scalability/security work. 6+ years shipping and scaling production platforms, comfort going deep with engineers on distributed-systems trade-offs, and the ability to turn ambiguity into clear plans. AI/ML or legal-tech background is a plus, not a must. Know someone great? Send them our way https://lnkd.in/gRXV_qiv
-
Peng Zhang liked thisPeng Zhang liked thisGreat piece by @JoannMuller at Axios, going behind the scenes of what we're building at Uber. Data is going to be the key unlock to accelerating safe, reliable autonomy everywhere - and we're just getting started. AV Labs data collection vehicles hit the road this month 🚀 https://lnkd.in/giyBg9gb
-
Peng Zhang liked thisPeng Zhang liked thisAgentic coding is moving fast, and GitHub Copilot Inline Suggestions continue to be used and loved by millions of developers. Excited to share how our team is making this experience even better with a specialized, low-latency model and post-training. Great work by Julia Gong, Ben Liggett and Ulugbek Abdullaev! Part 1 is out now, with more coming in Part 2. Stay tuned! We’re also hiring applied researchers to join us. Links are in the comments. #GitHubCopilot #VSCode
-
Peng Zhang liked thisPeng Zhang liked thisHi friends - We're continuing to hire for AI PM and Eng on the team, and wanted to do a plug for some additional roles below. If you know anyone that's a good fit, please send them over. Principle AI Engineers (L8 at most companies, US & India) - ping directly if interested Sr. Staff AI Engineers (L7 at most companies, US & India) - https://lnkd.in/g-cHHsBK L4 AI PM, US - https://lnkd.in/gBhneTDt L5 AI PM, Bangalore - https://lnkd.in/grJNcp6H I wrote a bit about the team/context in my earlier post (https://lnkd.in/p/gtMPfn-i), but TLDR - once-in-a-lifetime AI transformation, lots of really interesting tech innovation, systems-level thinking on steroids, and the team is great! One of the most fun I've had in AI because the results are so measurable and visceral -- strong foundations but with enough gaps that good people have the opportunity to directly drive outsized impact. Come join us! cc Benigne Shashaank Verena Anthony Raghav Surabhi Anuj Divya AditiSr. Staff Engineer (GenAI), San Francisco, United StatesSr. Staff Engineer (GenAI), San Francisco, United States
-
Peng Zhang liked thisPeng Zhang liked thisCall for Papers: AI Medicine; <https://lnkd.in/gQ7AQ7Fs>
-
Peng Zhang liked thisPeng Zhang liked this🚀 We’re hiring — come build the infrastructure powering the next generation of AI at ByteDance! Our teams in Singapore are growing, and we’re looking for talented engineers and builders across AI Infrastructure, ML Systems, Backend Engineering, SRE and ModelArk. If you’re excited about solving large-scale engineering problems and building systems that power real-world AI applications, we’d love to hear from you. 👀 🔥 Experienced Hires 🔹 Machine Learning Storage Infrastructure Engineer Build next-generation storage infrastructure for large-scale ML workloads. 👉 https://lnkd.in/gbScTF4P 🔹 Senior Backend Engineer — AML Engine Orchestration Work on large-scale orchestration systems powering ML/AI workloads. 👉 https://lnkd.in/dtEfJuUy 🔹 Site Reliability Engineer, Machine Learning Systems K8s × GPU × AI Infrastructure — build highly reliable systems at scale. 👉 https://lnkd.in/geA8VmnQ 🔹 Product Solution Architect — ModelArk (Singapore) Help customers and businesses unlock the potential of AI through ModelArk. 👉 https://lnkd.in/gpgjxbvY 🎓 Campus Hiring We’re also opening opportunities for 2027 campus talent across AI/ML Systems, Backend, AI Infrastructure, ML Engineering and more! 👉 Check out our campus opportunities: https://lnkd.in/gQF9z8sS Whether you’re an experienced engineer looking for your next challenge or a student ready to start your AI journey — come build the future with us in Singapore. 🇸🇬🤖 📩 Feel free to reach out to me / Xiaoci Li Yizhi Li if you’d like to learn more or explore which role might be the best fit for you. #ByteDance #Hiring #SingaporeJobs #AI #ArtificialIntelligence #MachineLearning #AIInfrastructure #MLSystems #BackendEngineering #SRE #Kubernetes #GPU #CampusHiring #TechJobsBackend Engineer - Machine Learning Storage Infra (Singapore) - Join ByteDanceBackend Engineer - Machine Learning Storage Infra (Singapore) - Join ByteDance
Experience & Education
Licenses & Certifications
Honors & Awards
-
Best Advanced AI Projects Award from Microsoft AI School
Microsoft AI School
Successful finished the PUE optimization project using deep neural network. Win the 1st place among 500+ AI project submitted to Microsoft AI School.
Languages
-
English
Full professional proficiency
-
Chinese
Full professional proficiency
Recommendations received
3 people have recommended Peng
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content