Combining Inverted and ANN Indexes for Efficient Hybrid Search at Scale

Explore the challenges and solutions for combining inverted indexes and Approximate Nearest Neighbor (ANN) indexes in modern search engines. Learn how to design an efficient system that leverages multiple indexes in parallel for hybrid search. Discover strategies for handling large RAM requirements, distributing ANN graphs across shards, updating vector embeddings quickly, and managing contention between indexing and vector search. Gain insights into the benefits of integrating traditional and new search approaches for improved results. This 38-minute conference talk, presented by Anubhav Bindlish at Haystack EU 2023, draws from his experience at Rockset and Meta Platforms (Facebook) in data indexing, query execution, and ML-based integrity infrastructure.