About
Summary
• Senior Software Engineer at LinkedIn with 7+ years leading backend…
• Senior Software Engineer at LinkedIn with 7+ years leading backend…
Activity
4K followers
-
Yan Xu reposted thisYan Xu reposted thisBig news! We are making it easier than ever to connect to opportunity on LinkedIn, so that creators can focus on doing what they do best (making stuff). Read more about our new Creator Marketplace, and all that we've been working on to help creators find success on LinkedIn.LinkedIn’s Creator Marketplace, a new ecosystem for brand/creator connection:LinkedIn’s Creator Marketplace, a new ecosystem for brand/creator connection:
-
Yan Xu reposted thisYan Xu reposted thisYour competitor can buy the same model you use. What they can't buy is the ground truth you've labeled in your own business's blood — and that, not the model, is the technical moat that's quietly compounding. A year of lessons on building it.
-
Yan Xu reposted thisYan Xu reposted thisThe model is rented. The eval is owned. One of the most useful frames I’ve seen for AI comes from Karpathy (https://lnkd.in/ghSRMD7F): Software 1.0 automated what we could specify. Software 2.0 automates what we can verify. That idea shows up everywhere in the AI stack. In reinforcement training, the verifier becomes the reward signal. In agent workflows, it checks whether the model picked the right tool and produced a usable result. In data and memory systems, it helps decide what to keep, fix, or forget. Different layer, same underlying question: Can we grade an attempt without a human in the loop? If yes, the system can improve quickly. If no, progress gets slower, more manual, and harder to trust. That’s why I think the real moat in AI is rarely the model itself. Models are increasingly accessible. What’s scarce is the domain-specific evaluator: the rubric, test set, and judgment loop that defines what “good” actually means. That’s the lens we use at Pax. In duty drawback, the hard part is not connecting an LLM to a customs database. The hard part is encoding broker-grade judgment: what counts as a valid import-to-export match, what is defensible in an audit, and what maximizes recoverable duty within a client’s risk tolerance. Pretty good leaves dollars on the floor. The model is rented. The eval is owned. #agent #paxai #paxiengineering #aiengineering
-
Yan Xu reposted thisYan Xu reposted thisI’ve long been impressed with what Anthropic is building. Every day, our engineers are using their tools to help build products that serve millions of people around the world. That's why I'm particularly excited that we're bringing Uber and Uber Eats into Claude. From today users can browse restaurants, check fare ranges, see ETAs and seamlessly complete trips or orders in the Uber and Uber Eats apps. It’s a great example of how we’re applying AI to everyday use cases, helping people travel and get what they need as seamlessly as possible. Delivering this required close collaboration across teams, bringing together expertise, real-time data platforms, and applied AI. A lot of care has gone into ensuring the experience is fast, reliable, and as intuitive as possible. Thank you to everyone who helped bring this to life. Excited to see it out in the wild! Uber Engineering
-
Yan Xu reposted thisYan Xu reposted thisI am saddened to share the news of Shuangyang Yang, a longtime employee at LinkedIn, who tragically took his own life yesterday. In this era of rapid AI advancements, knowledge workers, particularly software engineers, often face immense pressure. It is crucial for us to manage our mental health and maintain calm amidst work-related anxiety. Having worked at LinkedIn for nearly 10 years, I have had the privilege of collaborating with many talented engineers like Shuangyang. I invite everyone to join me in expressing condolences to Shuangyang and his family as they navigate this challenging time. For those who wish to support, there is a fund set up to assist them: https://lnkd.in/gx8x3Ke4Donate to Support for Shuangyang Yang’s Family, organized by Song LuDonate to Support for Shuangyang Yang’s Family, organized by Song Lu
-
Yan Xu reposted thisYan Xu reposted thisMy wife is 30 weeks pregnant. This morning Oracle laid her off by email. She didn't walk in off the street. She interned there twice. She earned her spot on that team. She went full time and never stopped giving more than what was asked. She built products, shipped code, created content for Oracle's own channels, mentored interns, and ran social evangelism for FreeSQL using her own audience. This morning she woke up locked out of every Oracle system she had access to. No call. No meeting. No manager. No warning. Just an email. And a DocuSign. That's it. That's how they did it. We have a 17 month old at home. And Oracle couldn't pick up the phone. I'm not writing this out of anger. I'm writing this because people need to see what this actually looks like from the inside. This is the human cost behind the headlines. She is one of the most talented and hardworking people I know and she deserved better than an email. If you are hiring in software engineering or technical product management, her name is Layla Massey and she is ready for what's next. #Oracle #TechLayoffs #OpenToWork
-
Yan Xu reposted thisYan Xu reposted thisShuangyang's funeral will be held on April 16th, from 10:00 AM to 12:00 PM, at Grissom's Chapel & Mortuary, 267 E. Lewelling Blvd, San Lorenzo, CA 94580.Donate to Support for Shuangyang Yang’s Family, organized by Song LuDonate to Support for Shuangyang Yang’s Family, organized by Song Lu
-
Yan Xu shared thisYan Xu shared thisThere’s a lot of anxiety right now about what AI will do to jobs. Will it automate work? Replace roles? Shrink opportunity? On the most recent episode of Building One, I asked Anton Osika, founder and CEO of Lovable, what people misunderstand most about the rise of AI-powered building tools. His answer resonated with me: “It’s not a fixed pie… new pies are being created.” When it becomes easier to build software, more people start companies. More products get created. And entirely new markets appear. Anton shared the story of a founder who built an AI healthcare staffing platform and hit $1M in revenue in five months. This would’ve been nearly impossible before without hiring a team of engineers. That’s the part I think we’re underestimating: more builders → more companies → more economic activity → more opportunities. It’s not just about automation. It’s about expanding what’s possible. ▶️ LinkedIn: https://lnkd.in/dJFNSNGE ▶️ Spotify: https://lnkd.in/dagJufKS ▶️ Apple: https://lnkd.in/dc4U7Ggz
-
Yan Xu liked thisYan Xu liked this[Building with & for AI] Introducing Memory Stargraph, an AI memory Visualizer for GBrain 👏 Memory Stargraph is a local web service for exploring a GBrain knowledge base through an interactive star-cloud entity graph. I built it because my local agent clusters now share the same GBrain-backed memory infrastructure across multiple hosts and agent runtimes. Instead of each agent carrying more context forever, they can rely on a shared, evolving knowledge graph that stays compact, searchable, and useful over time. Memory Stargraph makes that shared memory visible. You can see what agents have created in GBrain, inspect entities, explore relationships, create new nodes, connect them, modify links, attach media, and follow the map as it loads new neighborhoods on demand. It is part debugging tool, part knowledge browser, and part “wait, the agents remembered all this?” moment. It’s also surprisingly mesmerizing. Many thanks to Garry Tan for his GBrain work, and I am happy to be among the first serious users of #GBrain! If you’re interested in setting it up for your own agents: https://lnkd.in/dtRQxcWK #AI #AIAgents #KnowledgeGraph #LocalAI #OpenSource #BuildInPublic #Codex #OpenClaw
-
Yan Xu liked thisYan Xu liked thisI gave a few workshop what an AI-native org actually looks like lately, and wrote about what I shared in my newsletter. The puzzle in the room kept circling: usage charts are climbing, the spend is climbing -> but the business isn't moving as fast as the AI adoption picks up. When building becomes cheap, coordination becomes much more expensive than it seems. Most companies are parked at "AI-augmented" — tools layered on the old process. The leap to "AI-native" is a different architectural conversation than the one their adoption strategy is having. link to the article in the first comment 👇
-
Yan Xu liked thisYan Xu liked thisLast week marks my 11 year journey's end at LinkedIn. I am truly grateful for the opportunities the company have provided over the years. When I started, it was a small ads team, and I am glad to be part its incredible growth and evolution over the years. My favorite part was the people I've worked with together- so many folks I got to learn from and collaborate together to build something new, and I learned a lot from each and every one of them. For those impacted recently, my heart reaches out, and hope for for the best for you as well.
-
Yan Xu liked thisYan Xu liked thisThe Spirit of New York On June 10, we hosted an NBA Finals Game 4 watch party. When the Knicks fell behind by 29 points, we were basically hosting a group therapy session. People were staring into their drinks. A few were talking about heading home early. Most of us had already accepted the loss. Then, point by point, the Knicks fought their way back. Every basket got a bigger reaction than the last. Nobody left. Nobody checked their phones. The room went from resignation to belief. And then came the play I'll never forget. Brunson launched a three and missed. For a split second, the entire room groaned. Then OG crashed in for the tip-in. The place exploded. I've never seen a room go from heartbreak to euphoria that fast. A complete emotional roller coaster. By the end, complete strangers were celebrating together like old friends. We complain. We panic. We start drinking like losers when we're down 29. But somehow, we never stop believing a comeback is possible. This is New York. P.S. If you want an AA LAW jersey, DM me. 😉 #Knicks #NewYorkForever #NBAFinals #NewYork #Community #AALaw
-
Yan Xu liked thisYan Xu liked thisToday marks one year since I joined OpenAI 🎉 The most exciting part of this journey has been witnessing how profoundly coding agents have changed the day-to-day work of software engineers—especially since early 2026. For more than a quarter, I haven’t written a single line of code, or even attempted to. My three essential productivity tools are now the browser, Slack, and Codex. At the same time, there is enormous room to improve agent harnesses, orchestration, and how humans and agents work together. I can’t wait to see how my day-to-day continues to evolve in my second year at OpenAI.
-
Yan Xu liked thisYan Xu liked thisThrilled to share our team's work at Meta on autonomous agents for ML serving efficiency at @Scale AI + Data 2026. 💡 The idea: Inference is quickly becoming the next major bottleneck in AI infrastructure. As models multiply and grow, manual optimization just doesn't keep up anymore. We built an agent framework that does the heavy lifting - surfacing opportunities, proposing changes, and shipping them with human approval - composing specialized agents over a shared data foundation to dynamically optimize resource allocation across our serving fleet. 📈 The impact: This is in production today. Our agents have delivered tens of millions of dollars in infrastructure efficiency - reducing GPU and CPU footprint across hundreds of models that manual processes simply couldn't keep up with. 🤝 Huge shoutout to my brilliant teammates Angineh Keshishian, Bikash Sharma, Jinfu Leng, Zhihua (Mia) Gu, Michael Fulthorp and everyone at Meta who helped bring this to life. Great connecting with so many peers pushing the boundaries of AI and infrastructure. Excited to see how autonomous systems keep evolving. #Scale2026 #MachineLearning #AIInfrastructure #AutonomousAgents #MetaEngineering #META
-
Yan Xu liked thisYan Xu liked thisCopilot Cowork is now generally available worldwide, now with multi-model support! Every organization can put long-running agents to work on complex, multi-step tasks, grounded in your organization's unique knowledge and know-how. https://lnkd.in/g3gD9TKd
-
Yan Xu liked thisYan Xu liked thisExcited to share that I've joined ElevenLabs as a Deployment Strategist, helping enterprises build and deploy high quality agents at scale. Feeling super grateful for the opportunity and thrilled to help transform how people and businesses communicate with the world. Also wanted to give a special shoutout to all the friends I've made along the way at LinkedIn over the last 6 years. I would not be in this position without the lessons learned. Thank you to Jill, Rohan, Alice, Keren, Ryan, Lauri and Matthias for getting me here.
Experience & Education
-
NANKAI ALUMNI ASSOCIATION INC
View Yan’s full experience
See their title, tenure and more.
or
Licenses & Certifications
-
Advanced Learning Algorithms
DeepLearning.AI, Stanford University
-
Machine Learning Specialization
DeepLearning.AI, Stanford University
-
Unsupervised Learning, Recommenders, Reinforcement Learning
DeepLearning.AI, Stanford University
-
Supervised Machine Learning: Regression and Classification
DeepLearning.AI, Stanford University
-
-
-
-
-
Parallel Programming in Java
Coursera Course Certificates
Courses
-
Algorithm
-
-
Computer System
-
-
Computer Vision
-
-
Data Mining
-
-
Discrete and Data Structures
-
-
Functional Programming
-
-
Introduction to Database
-
-
Mathematical Modeling
-
-
Object Oriented Design and Analysis
-
-
System Programming
-
-
Web Development
-
Projects
-
AlgoFun - An algorithm portal for geeks(MEAN Stack Web)
-
See project• Constructed a fully featured blogging web app for people to share and discuss innovative ideas for algorithms
• Designed and implemented an interactive front-end UI utilizing Angular, Bootstrap, HTML, and CSS
• Created a RESTful API web server with Node.js, Express and persisted data in a MongoDB database
• Applied PassportJS and JWT for user authentication increasing security while maintaining usability -
Markdown Document Processor(Java)
-
• Developed a markdown document processor that handles headers, numbering of headers as well as itemized and enumerated lists
• Provided user-friendly command line interface to process a series of documents
• Performed unit testing using JUnit framework to ensure the full functionality of the application -
File Search Engine(C/C++)
-
• Developed a file search engine, searching files containing all of the words in the query
• Implemented linked-list, hash-tables, inverted index, and index files for efficient query processing
• Built a multithreaded web server front-end to the query processor using POSIX Socket API and Pthreads -
Space Invader Game(Racket)
-
• Designed and developed a space invader game utilizing functional programming design principle
• Applied generic data definitions and predefined Higher Order Functions provided by Racket(map, filter, etc) for less code, less redundancy
• Unit testing to ensure full functionality -
Algorithms Course Projects(java)
-
• Built a computational model to estimate the percolation threshold using union-find algorithm
• Implemented an efficient, sorting-based algorithm to find every line segment that connects a subset of 4 or more of the points in the plane
• Designed a data type to represent a set of points in a unit square using 2d-tree to support efficient range search and nearest neighbor search
• Developed an 8-puzzle solver using the A* search algorithm
Languages
-
English
-
-
Chinese
Native or bilingual proficiency
Other similar profiles
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content