Unveiling the Secrets of LinkedIn’s Massive People Search Expansion: A Journey to 1.3 Billion Users

LinkedIn has finally launched its highly anticipated AI-powered people search feature this week, following a long period of development. The launch of this new tool comes three years after the introduction of ChatGPT and six months after the release of LinkedIn’s AI job search feature. The delay in launching the people search tool highlights the challenges of implementing generative AI on a large scale, especially for a platform with 1.3 billion users.

In exclusive interviews with the LinkedIn product and engineering team responsible for the launch, it was revealed that the new AI-powered people search feature revolutionizes the search experience on the platform. Unlike the previous keyword-based search system, the new AI-powered search understands natural language queries and semantic meaning. For example, a user can now search for “Who is knowledgeable about curing cancer?” and the system will provide relevant results related to oncology and genomics research, even if the word “cancer” is not explicitly mentioned in the profiles.

The new system not only focuses on relevance but also takes into account the user’s network, prioritizing connections that can serve as a bridge to experts in the field. This approach enhances the user experience and improves the quality of search results.

LinkedIn’s journey in developing the AI-powered people search feature offers valuable lessons for enterprise practitioners. The company’s approach of focusing on one vertical first, as seen with the success of its AI job search feature, provided a blueprint for scaling up to a billion-user product. By following a replicable pipeline of distillation, co-design, and optimization, LinkedIn was able to overcome the challenges of deploying generative AI at scale.

One of the key challenges faced by the team was training models to balance policy adherence with user engagement signals. The breakthrough came when they adopted a multi-teacher ensemble approach, which improved the relevance and efficiency of the system. The team also had to optimize the architecture for efficient retrieval and ranking, leading to a 10x increase in throughput.

LinkedIn’s pragmatic approach to building recommender systems over chasing “agentic hype” underscores the importance of focusing on practical tools rather than flashy AI agents. The company’s emphasis on perfecting the recommender system first has proven to be a strategic advantage in delivering relevant search results to users.

Overall, LinkedIn’s playbook for enterprise AI development emphasizes pragmatism, codifying processes, and relentless optimization. By mastering the pipeline of co-design, distillation, and creative optimizations, enterprises can achieve success in deploying AI-powered solutions at scale.

For enterprises embarking on their AI roadmap, LinkedIn’s journey serves as a valuable guide for navigating the complexities of implementing generative AI in real-world settings. By prioritizing practical tools and optimization strategies, companies can deliver enhanced user experiences and drive innovation in AI technology.