Netflix evolved their Graph Search platform to support natural language queries by integrating LLMs. The system converts user questions into structured DSL filter statements through a multi-stage process: RAG-based context engineering to identify relevant fields and controlled vocabulary values, LLM-based generation with carefully crafted instructions, and deterministic validation for syntactic and semantic correctness. Key innovations include field and vocabulary RAG to manage context size, UI visualization of generated filters as chips and facets, and @mention functionality for explicit entity selection. This approach bridges the gap between complex federated graph queries and intuitive user intent while maintaining trust through transparency.
Table of contents
The Need for Intuitive Search: Addressing Business and Product DemandsUnder the Hood: Our Approach to Text-to-QueryContext EngineeringGet Netflix Technology Blog ’s stories in your inboxThe InstructionsValidationBuilding ConfidenceEnd-to-end architectureSummaryCreditsSort: