An introduction to the user embedding, with learnings inspired by the TikTok algorithm

​What if 64 numbers could determine who you are, who you should date, your life trajectory, and whether or not you'll get diabetes? Could you use this information to make your life better?

​Embeddings represent the inner learnings of neural networks. Painstakingly etched into existence over time, their precise values enable more accurate predictions and help gauge similarity across items, enabling efficient information retrieval.

​While you've likely worked extensively with text or image embeddings already, you might be less familiar with User Embeddings - the learned representations of users and how they interact with other users and items. Compared to text and images, user-related data has historically been much more difficult to come by as nearly all of it (including yours!) resides in inaccessible digital silos, locked away by large internet companies.

​As AI models become more ubiquitous in their abilities and outputs, Context about the user becomes more important than ever. The user embedding will power AI agents to create hyper-personalized life service recommendations.

​In today's talk, we'll introduce the user embedding - how it's created, what it represents, and how to unlock and pair it with agents to shape the future of humanity and agentic AI.

​About the Speaker:
Drew Kirchhoff
x: https://x.com/_drewkirchhoff
LinkedIn: https://www.linkedin.com/in/drew-kirchhoff/
Current: Founder at ShareDot
Previous: Algo PM & DS at TikTok