Our Early Journey to Transform Instacart’s Discovery Recommendations with LLMs

What this post added

This post details the early journey of Instacart in transforming its discovery recommendations using LLMs. It outlines the limitations of traditional recommendation engines, such as difficulty in scaling personalized content and lack of cohesion. The proposed solution involves an AI-native platform with a top-down, cascaded generative approach. Key phases include page design & theme generation, retrieval keyword generation, quality and diversity filtering, and product and pagewise ranking. The methodology emphasizes personalized themes, retrieval-compatible descriptors, and a structured schema for interpretability.