LinkedIn's AI search for 1.3 billion users: a disciplined approach to AI

Really exciting insights into LinkedIn's journey to applied enterprise AI through mastering the pipeline. This is not an easy problem to solve, as we've done variations on this at exchange.design (on a comparatively tiny scale) for partners looking at resumes, past performance, and other business development needs. The value of our approach has been technology agnostic and suited to the problem, which seems to be exactly aligned with LinkedIn: "The real value for enterprises today lies in perfecting recommender systems, not in chasing "agentic hype." He also refused to talk about the specific models that the company used for the searches, suggesting it almost doesn't matter. The company selects models based on which one it finds the most efficient for the task." https://lnkd.in/eaygzJZ5

🚀 LinkedIn just dropped a masterclass in scaling generative AI — and it’s pure gold for anyone building AI products at scale. Here’s the recipe straight from their AI “cookbook”: 1️⃣ Start narrow. Scale wide. LinkedIn didn’t go big on day one. They nailed AI-driven job search first — then cloned that success for people search. Precision beats ambition. 2️⃣ Systemize the magic. Their pipeline turns chaos into clarity — from “golden” training sets to distilled models (7B → 1.7B → 220M) to lightning-fast indexing. It’s not luck. It’s process. 3️⃣ Skip the sizzle. Serve the steak. While others chase shiny AI agents, LinkedIn doubled down on retrieval, ranking, and real results. Pragmatism wins. Every. Single. Time. Bottom line? The future of AI belongs to builders who prioritize clarity over complexity and execution over excitement. #AI #GenerativeAI #AIBuilders #TechLeadership #ProductInnovation #LinkedInAI VentureBeat Matt Marshall https://lnkd.in/g3VA6V5x

𝗜𝗻𝘀𝗶𝗱𝗲 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻’𝘀 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜 𝗰𝗼𝗼𝗸𝗯𝗼𝗼𝗸: 𝗛𝗼𝘄 𝗶𝘁 𝘀𝗰𝗮𝗹𝗲𝗱 𝗽𝗲𝗼𝗽𝗹𝗲 𝘀𝗲𝗮𝗿𝗰𝗵 𝘁𝗼 𝟭.𝟯 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 𝘂𝘀𝗲𝗿𝘀 https://lnkd.in/e4h8w4SU LinkedIn is launching its new AI-powered people search this week, after what seems like a very long wait for what should have been a natural offering for generative AI. It comes a full three years after the launch of ChatGPT and six months after LinkedIn launched its AI job search offering. For technical leaders, this timeline illustrates a key enterprise lesson: Deploying generative AI in real enterprise settings is challenging, especially at a scale of 1.3 billion users. It’s a slow, brutal process of pragmatic optimization.

LinkedIn is launching its new AI-powered people search this week, after what seems like a very long wait for what should have been a natural offering for generative AI. It comes a full three years after the launch of ChatGPT and six months after LinkedIn launched its AI job search offering. For technical leaders, this timeline illustrates a key enterprise lesson: Deploying generative AI in real enterprise settings is challenging, especially at a scale of 1.3 billion users. It’s a slow, brutal process of pragmatic optimization. The following account is based on several exclusive interviews with the LinkedIn product and engineering team behind the launch. https://lnkd.in/geQmv5wi #LinkedIn #GenerativeAICookBook #HowItScaledPeopleSearch

Deploying generative AI for 1.3 billion users isn't plug-and-play. It's a "slow, brutal process of pragmatic optimization." LinkedIn just launched its AI People Search, deploying it across a 1.3B member graph. In the rush to integrate generative AI, it's easy to underestimate the sheer engineering lift required for enterprise-scale deployment. LinkedIn's recent overhaul of its People Search is a masterclass in what it truly takes. While the new natural language search capability is a massive win for users, the story behind it is one of immense technical challenge and strategic patience. A VentureBeat article peels back the curtain, revealing that this wasn't a quick project. The team described it as a "slow, brutal process of pragmatic optimization" to serve their 1.3 billion members effectively (https://lnkd.in/gRqWyGx8). This is a critical lesson for tech leaders. The leap from a proof-of-concept to a reliable, scalable, and valuable enterprise feature is massive. It involves building a proprietary "cookbook" – a set of methodologies and optimized models tailored to the specific needs of the platform. This wasn't about simply integrating an off-the-shelf API. It was about deep, foundational work to ensure the AI delivered measurable value without compromising performance or accuracy at a global scale. It’s a powerful reminder that true innovation in AI requires not just brilliant algorithms, but also world-class engineering discipline. 🔥 Enterprise AI deployment is a marathon, not a sprint. 💡 Scaling AI requires deep, pragmatic optimization. 🤖 Proprietary AI 'cookbooks' are a strategic advantage. 📈 True innovation demands engineering excellence. 👇 I’d love to hear thoughts and takeaways—drop them in the comments. #AIstrategy #TechLeadership #Engineering #GenerativeAI #EnterpriseTech

LinkedIn is launching its new AI-powered people search this week, after what seems like a very long wait for what should have been a natural offering for generative AI. It comes a full three years after the launch of ChatGPT and six months after LinkedIn launched its AI job search offering. For technical leaders, this timeline illustrates a key enterprise lesson: Deploying generative AI in real enterprise settings is challenging, especially at a scale of 1.3 billion users. It’s a slow, brutal process of pragmatic optimization. #linkedin #enterpriseai #aipowered #aisearch #ai #genai #aidevelopment #aideployment #aiadoption #aioptimization #largeenterprises #scalingai #aitransformation #llms #ml #geo #linkedinsearch #casestudy #lessonslearned

🧠 Inside LinkedIn’s Generative AI Cookbook: Scaling People Search to 1.3 Billion Users LinkedIn has officially launched its new AI-powered people search — a major leap from traditional keyword-based search to a semantic, intent-driven system powered by large language models (LLMs). What’s new? Instead of searching for exact terms like “cancer,” users can now ask natural questions like “Who is knowledgeable about curing cancer?” The AI understands context and surfaces relevant experts — even if their profiles don’t use the exact keywords. Why did it take so long? Despite the hype around generative AI, LinkedIn’s journey shows that deploying AI at enterprise scale is a slow, complex process. It took three years post-ChatGPT and six months after launching AI job search to get here. The real story is the “cookbook” — a repeatable pipeline of: • Policy-driven model distillation • Co-design between product and engineering • Relentless optimization (including pruning and reinforcement learning-based summarization) Key technical highlights: • A 7B parameter policy model distilled into a 220M student model with <1% relevance loss • 20x input size reduction via an RL-trained summarizer • 10x throughput gain in ranking • GPU-based retrieval infrastructure to handle the 1.3B-member graph This isn’t about flashy agents — it’s about building robust tools. As LinkedIn’s VP of Product Engineering Erran Berger puts it: “Agentic products are only as good as the tools they use.” For enterprises, the takeaway is clear: 1. Be pragmatic — win one vertical first 2. Codify your process — build a repeatable pipeline 3. Optimize relentlessly — the real gains come after the first model LinkedIn’s approach is a masterclass in real-world AI deployment at scale. Join Artificial Intelligence School to learn how to build and deploy enterprise-grade AI systems. Our expert-led programs are designed for professionals who want to go beyond the hype and master the fundamentals. #aischool #artificialintelligenceschool #generativeai #enterpriseAI #llm #machinelearning #searchtechnology #recommendersystems #aiengineering #linkedinAI #productengineering

OpenAI Just Gave Every Company a Collective Brain. Today, OpenAI dropped what might be the most powerful enterprise update yet it's called Company Knowledge, and it might just redefine how businesses think, work, and remember. Here's the wild part ChatGPT can now connect directly to your internal tools like Slack, Google Drive, GitHub, Notion, and HubSpot, and pull knowledge from them in real-time. Imagine this Instead of searching 10 folders, scrolling through old Slack threads, or asking your manager for that lost doc you can just ask ChatGPT: Summarize everything our team discussed about Q4 marketing strategy from Slack and Drive. and it answers, with citations, clarity, and context. That's not just a productivity hack. That's organizational intelligence. Here's why this is bigger than it sounds It's powered by a specialized GPT-5 model that can analyze multiple apps simultaneously. It respects user permissions, meaning no creepy overreach. It turns scattered data into one unified memory your company's own neural network. In psychology, we call this distributed cognition the idea that thinking doesn't just happen in our heads… it happens across systems, tools, and people. OpenAI just gave that concept a literal interface. Why It Matters This update isn't just about efficiency. It's about context, the one thing AI has always struggled with. When machines understand not just your query, but your company's ecosystem, you move from "AI assistant" → to "AI collaborator." We're watching AI evolve from talking back → to thinking with. And in the bigger picture Microsoft's Copilot, Google's Gemini Enterprise, and now OpenAI's Company Knowledge, they're all racing to own the cognitive layer of work. The place where information turns into decisions. I think the real question isn't who wins the AI race it's which companies learn to think with their AI first. Because in this new era, your biggest competitive advantage won't be how much your team knows but how fast your team can remember, connect, and act. What do you think, will AI knowledge systems replace how we "search" inside companies, or will they reshape how we collaborate entirely? #OpenAI

#Latestissuealert #AIUngeeked #newsletter AI in the deck or at the desk? is the latest issue at AI Ungeeked - The newsletter that talks AI in plain speak. This quick- read issue is all about: - Enterprise AI adoption (or lack of) - the employee perspective - Practical use of AI chatbots in speeding up hiring without being evil - Big tech's investment in improving AI literacy at large, and, - Atlas launched: The rise of Agentic browsing and why you might want to shrug it for now, but also safely experiment with it :) All this and more at: https://lnkd.in/dJSuKaaT with Suchitra Laxman Kavitha Murali <This is a personal sub-stack and does not necessarily reflect the opinion of my employer>

We’ve all been there. You know that document exists somewhere. But where? And before you know it, you’ve wasted your morning 🤦 That’s the problem Microsoft’s new Knowledge Agent aims to fix 💡 Now in public preview, it uses AI to understand context, not just keywords, so employees can find the right information faster, right where they’re working. But that’s not all. It can help organize your content, identify and clear outdated pages, suggest content gaps that need to be filled, and even help create pages 💪 Get the full picture on Knowledge Agents ⤵️ https://lnkd.in/eKfvMBMr #Microsoft365 #SharePoint #KnowledgeAgent #AI #DigitalWorkplace #Cloudwell