Session brief
Building LinkedIn's GenAI Platform — Xiaofeng Wang
Overview
This talk details LinkedIn's journey in building a Generative AI platform, emphasizing its critical role in supporting evolving AI product experiences. The platform evolved from simple prompt-based features to sophisticated multi-agent systems, necessitating a unified infrastructure to manage complexity, enforce best practices, and bridge the gap between AI and product engineering.
Who should watch
- AI Engineers
- Product Managers
- Builders of AI systems and platforms
- Those looking to integrate AI into existing products
- Teams struggling with the complexity of AI development and deployment
Key takeaways
- LinkedIn's GenAI platform evolved through distinct stages: initial prompt-based features, a second generation with personalized recommendations using RAG, and finally, a multi-agent system for recruiters.
- A unified Python SDK built on a popular launching framework became central to orchestrating LLM calls and integrating with large-scale infrastructure.
- Key platform components include prompt management for version control and operational stability, conversational and experiential memory for context and knowledge inference, and a skill registry for easy API invocation by agents.
- The platform aims to serve as a unified interface, abstracting the complexity of the underlying AI ecosystem for developers.
- Building a dedicated GenAI platform is crucial because GenAI systems are compound AI systems where the lines between optimization and serving phases blur, requiring a different approach than traditional AI.
- Hiring for such a platform prioritizes strong software engineering skills, potential over experience, and a diversified team that can learn and adapt.
- Python is strongly recommended as the primary technology stack due to its prevalence in research and open-source communities, and its scalability.
- Focusing on developer experience and aligning the platform with existing workflows is key to successful adoption.
Notable quotes
*GenAI systems is a compound AI system.*
*The short answer is yes [we need this platform].*
*GenAI is totally different new AI systems compared to the traditional AI systems.*
Unofficial community note. Prefer the recording for nuance.