Building Our New Personalized AI-Powered Premium Experience

November 1, 2023

(This article originally appeared on LinkedIn)

Advancements in AI are already starting to transform jobs, skills, and career paths, and we know that our members and customers are thinking about the implications for this new world of work. That’s why I’m excited to share that we’re rolling out a new AI-powered LinkedIn Premium experience, starting with a selected group of US based Premium subscribers today. These members will see AI-powered takeaways on the platform to help them unlock hidden opportunities, from how to get more knowledgeable about important topics, to what actions to take to advance their career.  

These features are powered by generative AI (GAI) and deliver highly personalized and relevant takeaways in seconds through a conversational experience. Whether navigating career changes, building a business, learning a new skill, or crafting their voice, our new AI-powered LinkedIn Premium experience is a partner to help our subscribers stay ahead.

To engineer this experience, we built new infrastructures and foundations to support rapid GAI-related innovation and leveraged our existing technology to deliver more value to Premium members in a personalized and unique, LinkedIn way. We also made substantial investments in our prompt engineering and evaluation framework to ensure that the conversations are trustworthy and professional. 

Because we wanted to get this product to market quickly, we rallied our product and engineering teams to innovate together with even greater urgency to build products that deliver a new level of personalization and intelligence for our members. Instead of striving for perfection, we encouraged our teams to operate with a mindset that optimizes for agility, adaptability, and continuous learning. 

Building a new technical foundation for innovation

At LinkedIn, AI powers every product we build and every experience we deliver, so working with AI isn’t new for our teams, but working with generative AI requires a different approach

Roughly a year ago, our engineering team started building a new foundation to support GAI-related development from end-to-end so that our teams could safely explore and build with this new, rapidly evolving technology. With centralized access to tools and knowledge to support exploration with GAI, our AI and product engineering teams have been able to collaborate and innovate at an impressive pace. Recruiter 2024, which we released as a pilot in October, is a great example of a product we built and launched from end-to-end via our GAI infrastructure. 

This foundation enables us to build and design new member and customer experiences on top of a shared set of components that are pluggable and extensible. A separate orchestration layer – which we built using a framework we formed for blending results together in a low latency and high quality way – executes prompts and helps us process responses from large-language-models (LLMs) incrementally in a streaming fashion. The end result for members is personalized responses that are not only conversational but also leverage insights that are pulled from the entire LinkedIn ecosystem.

All of this helps us scale GAI to many different use cases, reducing the cost and complexity of innovation while delivering new, valuable experiences faster to our members and customers.

Delivering a highly personalized and responsive experience 

With our newest AI-powered LinkedIn Premium experience, we set out to deliver experiences that feel relevant and supportive to each member’s career journey and professional needs. With that goal in mind, applying an off-the-shelf LLM technology was far from sufficient. We not only built our own tooling and platforms to utilize the capabilities of existing OpenAI models in our tech stack, but also leveraged the knowledge we have about our members’ professional skills and interests, as well as insights across LinkedIn, in order to create a personalized experience for our members in a unique, LinkedIn way. 

Additionally, we also need to ensure that the interactions are responsive and seamless to deliver value and keep members engaged. We are able to provide personalized takeaways and insights within roughly five seconds. To pull that off, our engineering team had to find the right balance between personalization and response speed, and built various mechanisms such as streaming responses, offline data processing, and pre-computation to reduce real-time response latency.

Engineering natural conversations that can be trusted

Creating GAI based conversations that are natural and trustworthy is as crucial to the experience as making them personalized and responsive. However, achieving this is challenging.

GAI is excellent at synthesizing results and creating natural sounding responses – but left unchecked, it can also produce bad responses that can have an inappropriate tone or content (such outputs are often referred to as “hallucinations”). 

To tackle this challenge, we invested a great deal of time and effort learning and iterating how to build high-quality interactions that minimize bad or trust-violating responses. We built a set of tools and processes to evaluate the quality of responses and hallucination rates, which guides our teams to iterate on prompt engineering accordingly. During the evaluation, we also analyze the tone of responses to make sure they are respectful and professional. Finally, the team also adds trust checks to establish safety boundaries and performs thorough red teaming trust exercises regularly. 

Our ongoing commitment to our members

As the world of work continues to change rapidly amid the new advancements in AI, we're thrilled to introduce this new experience and put GAI in the hands of our Premium subscribers, enabling them to more effectively navigate their careers and achieve their professional ambitions. 

But this is just the beginning of our exciting journey. Looking ahead, we are committed to responsibly designing, developing, testing, and refining our technologies and products to unlock the full potential of GAI and help our members and customers discover new opportunities and advance their career paths and businesses – one connection, conversation, learning session, or engagement at a time.