The AI Evolution of Graph Search at Netflix: From Queries to Natural Language (netflixtechblog.com)

Netflix has announced a significant evolution in its Graph Search platform, transitioning from structured queries reliant on complex languages to a more user-friendly natural language interface. Leveraging advancements in AI, particularly Large Language Models (LLMs), Netflix aims to simplify the search experience for users by allowing them to input queries in everyday language. This shift is crucial as it addresses usability challenges within a system that often confuses users with its technical jargon and complex interfaces, promoting more efficient data retrieval across its vast enterprise ecosystem. The implementation involves generating Graph Search Filter statements directly from natural language inputs, ensuring that these queries maintain syntactic, semantic, and pragmatic correctness. Key to this process is a method of context engineering that draws on metadata from GraphQL queries to inform the LLM, streamlining the conversion from user phrases like "I want to see all movies from the 90s about robots from the US" into structured queries. Netflix's approach involves sophisticated retrieval-augmented generation (RAG) techniques to curate relevant fields and vocabulary for the LLM, enhancing accuracy and minimizing latency. This development not only improves the user experience but also sets a foundation for advanced AI applications in data retrieval, with potential implications for industries relying on complex data interactions.