Session Description

This session will cover the architecture and engineering decisions behind modernizing search at Netflix, including:

• Optimizing Joint-BERT models for real-time inference to achieve low latency response times.
• Designing a hybrid engine that combines lexical search with both per-field and broad document level embeddings to handle long-tail exploratory queries.
• Managing the lifecycle of search models, including annotation strategies, labeling consensus, and repeatable benchmark workflows.
• Real-world results and business impact: Lessons learned from A/B testing, holdback experiments, and their effect on streaming lift and engagement.