Milvus 3.0: A More Powerful Retrieval Engine | Milvus Meetup SF

At the Milvus Meetup in San Francisco, Jiang Chen (Head of Developer Relations and Solutions at Milvus/Zilliz) presented “Milvus 3.0: A More Powerful Retrieval Engine”. Milvus is expanding beyond traditional pure vector search into a comprehensive retrieval engine—combining dense vector, full-text keyword, and structured search directly over data stored in open lake formats. In this session, Jiang unpacks why modern AI retrieval workloads require more than simple approximate nearest neighbor (ANN) search as datasets scale into hundreds of billions of records. He walks through the three architectural pillars of Milvus 3.0: advanced retrieval capabilities (custom dictionaries, server-side multi-stage reranking, and SQL-like grouping/aggregation), native data modeling with StructArray for nested multimodality and in-place schema changes, and a lake-native storage foundation integrating with open table formats like Parquet, Lance, and Iceberg. What you'll learn: The evolution of Milvus from a vector database to a lake-native retrieval engine Milvus 3.0’s three architectural pillars: Retrieval, Data Modeling, and Storage Advanced retrieval: custom dictionaries, full-text search, and server-side multi-stage reranking pipelines SQL-like capabilities in vector search: ORDER BY, grouping, and aggregations Modeling multimodal documents, hierarchical data, and video frames natively with StructArray In-place schema evolution and support for nullable vectors Lake-native External Collections: querying open table formats (Parquet, Lance, Iceberg) with zero-copy indexing Fast metadata snapshotting for reliable backups across billion-scale vector datasets Chapters: 0:00 Introduction & Billion-Scale Vector Challenges 1:05 The Rise of Vector Lakes 2:26 Lake-Native Architecture & Open Table Formats 3:15 Pillar 1: Richer Retrieval Beyond Pure ANN 4:34 Custom Dictionaries for Full-Text Search 5:44 Server-Side Multi-Stage Reranking (MapReduce Style) 9:08 SQL-Like ORDER BY & Distributed Aggregations 11:46 Demo: Multi-Signal E-Commerce Reranking 13:30 Pillar 2: Native Data Modeling & StructArray 14:10 Modeling Multimodal Frames & Hierarchical Documents 16:03 Autonomous Driving Use Case & Schema 17:00 Demo: Frame-Level Hybrid Video Search 18:00 In-Place Schema Evolution & Null Vectors 20:45 Pillar 3: Storage Foundation & Open Table Formats 22:15 External Collections & Zero-Copy Indexing 23:25 Fast Metadata Snapshots for Billion-Scale Backups 24:56 Spark Connector & Ecosystem Integration 25:17 Milvus 3.0 Availability & Milvus Cloud Preview This talk was recorded at the Milvus Meetup in San Francisco (525 Market St). The event brought together search, ML infrastructure, and data platform engineers to explore the algorithms, infrastructure, and production patterns behind modern AI-powered retrieval systems. Watch the full event recording: Milvus Meetup San Francisco October 2026 (Full Event): https://youtu.be/itISrq5J-Ik Watch other talks from this meetup: Large Scale Entity Resolution with Milvus at 6sense: https://youtu.be/L7DhSBXXomo From Retrieval to Action: Building Production Agents on AWS: https://youtu.be/wmbsMV87AQQ Find us on: Zilliz Website: https://zilliz.com Milvus Website: https://milvus.io Zilliz LinkedIn: https://www.linkedin.com/company/zilliz/ Milvus LinkedIn: https://linkedin.com/company/67143129 Milvus X: https://x.com/milvusio Milvus GitHub: https://github.com/milvus-io/milvus Try Zilliz Cloud: https://cloud.zilliz.com/signup