Generative AI in the Real World: Faye Zhang on Using AI to Improve Discovery

Why It Matters

By turning semantic AI breakthroughs into production‑ready discovery and recommendation systems, companies can capture billions of dollars of otherwise lost revenue and deliver more personalized experiences at scale.

Key Takeaways

• PinLanding tackles $6.5T discovery gap in e‑commerce via semantic search.
• Vision‑language models achieve 99.7% top‑10 recall on fashion benchmark.
• Modern recommenders shift from correlation to causal, semantic reasoning.
• Knowledge distillation enables fast, high‑precision models for production use.
• Context engineering balances window size, efficiency, and structured data integration.

Summary

The podcast features Faye Zhang, a staff AI engineer at Pinterest, discussing how generative AI is being deployed to solve the massive "discovery crisis" in online retail. She explains PinLanding, a system that semantically organizes billions of catalog items to match user intent, aiming to recover trillions of dollars of missed sales.

Zhang highlights that leveraging vision‑language models and OpenAI’s CLIP has pushed recall on the public Fashion200K benchmark to 99.7% for the top‑10 results. She also notes a broader industry shift: recommendation engines are moving from simple correlation toward causal, semantic reasoning, addressing cold‑start problems and unifying intelligence across platforms like Netflix’s UniCore and YouTube’s Gemini‑powered recommendations.

Illustrative examples include a bride asking ChatGPT for a specific dress that never surfaces in catalog searches, Netflix’s shared‑representation architecture, and a knowledge‑distillation pipeline where GPT‑4’s precision is transferred to a lightweight model achieving millisecond latency. Zhang stresses the role of rigorous A/B testing platforms and UX design in turning algorithmic gains into real revenue.

The conversation underscores that enterprises must invest in foundation models, robust evaluation, and context‑engineering practices to unlock hidden value, improve user experience, and stay competitive in a market where discoverability directly translates to profit.

Original Description

AI engineer Faye Zhang joins Ben to discuss discoverability: how to use AI to build search and recommendation engines that actually find what you want. Listen in to learn how AI goes way beyond simple collaborative filtering—pulling in many different kinds of data and metadata, including images and voice, to get a much better picture of what any object is and whether or not it’s something the user would want.

Follow O'Reilly on:

LinkedIn: https://www.linkedin.com/company/oreilly/

Facebook: http://facebook.com/OReilly

Instagram: https://www.instagram.com/oreillymedia

BlueSky: https://bsky.app/profile/oreilly.bsky.social