Method

HNSW

Hierarchical navigable small-world graph index for nearest-neighbor search.

18 cases from 16 companies apply HNSW. 6 were checked in an online A/B test; 12 report results in production.

Typical tasks: Retrieve and rank from millions of items, Rank search results, Build a data platform, Learn user and item embeddings and 7 more. Companies: Yandex, Netflix, LY Corporation, Pinterest, CFT, Nextdoor and 10 more. Often combined with: Two-tower model, Contrastive learning, Vector / semantic search, DSSM, LLM-as-a-judge and 51 more.

Building the AI Retrieval Infrastructure Behind 20 Billion+ Vectors at HubSpotsaas-b2b2026-06
From Zero to Agents: Growing a Multi-Agent LLM Platform Inside a Companygeneral2026-05
Synchronizing the Senses: Powering Multimodal Intelligence for Video Searchmedia-streaming2026-04
Dense Retrievalmarketplace2025-11
Finding the Right Organization in a Photo, or Removing Kul Sharif from the Kazan Cat Gallerygeneral2025-11
Milvus: Building a large-scale vector DB for LINE VOOM's real-time recommendation systemsocial2025-07
How Large Language Models Help Users Formulate Queries in Yandex Mapssearch-engine2025-05
Establishing a Large Scale Learned Retrieval System at Pinterestsocial2025-01
How we improved the 'Unfamiliar' setting recommendation algorithm in Yandex Musicmedia-streaming2024-05
Shazam for Full Names: Finding a Match When a Name Can Be Written 20 Different Waysfintech2024-05
Adopting the Vespa search engine for serving personalized second-hand fashion recommendations at Vintedmarketplace2023-10
From Pre-trained to Fine-tuned: Nextdoor’s Path to Effective Embedding Applicationssocial2023-09
Embedding-based Retrieval with Two-Tower Models in Spotlightsocial2023-06
Semantic Search for Organizationssearch-engine2023-05
NER for Marketplace Search Queriesmarketplace2023-05
eBay’s Blazingly Fast Billion-Scale Vector Similarity Enginemarketplace2023-05
When a picture is worth more than wordstravel2022-12
Matching Session at the Data Fusion Conferencebanking2022-06