Webinar | One Document, Many Vectors: Entity- and Element-Level Search with Milvus StructArray

Searching a passage and returning a useful document are different tasks. Applications often need to connect a matching chunk to its parent article, show surrounding context, and avoid returning the same document multiple times. In this live webinar, Rohit Nijhawan, Senior Solutions Architect at Zilliz, and Philip Chen, Developer Advocate at Zilliz, explore how StructArray in Milvus keeps chunks, their embeddings, and metadata within each document entity to solve the challenges of traditional chunk-based RAG. If you are building production RAG, semantic document search, video search, question answering with citations, or multi-chunk retrieval pipelines, this session explains how StructArray unifies entity-level and element-level search without client-side stitching overhead. What you'll learn: ✦ Model Documents with Milvus StructArray: Organize chunk embeddings and metadata within each document, with the schema and indexes needed for search ✦ Explore a Document-Search Demo: Compare title and passage searches, locate matching chunks, filter array elements, and group results by document ✦ Build RAG with Context and Citations: Assemble matching passages and neighboring chunks to generate cited answers ✦ Cross-Modal Applications: See how the parent-and-element model extends naturally to video and audio segment search ✦ Reuse Existing Embeddings: Migrate chunk collections without re-embedding them, and understand when a flat collection remains the better choice ✦ Key Architectural Trade-offs: Memory considerations, schema evolution, and performance trade-offs for StructArray fields ✦ Comprehensive Live Q&A: Deep dive into PDF ingestion, legal citation modeling, technical diagram handling, and multidimensional embeddings Slides from this webinar are available here: https://24054828.fs1.hubspotusercontent-na1.net/hubfs/24054828/Webinars/Webinar%20_%20One%20Document%2c%20Many%20Vectors_%20Entity%20and%20Element-Level%20Search%20with%20Milvus%20StructArray.pdf Chapters: 00:00 Welcome & Speaker Introductions 00:41 Overview: Entity- and Element-Level Search with StructArray 02:45 The Challenge: Chunk-Based Retrieval vs. Returning Useful Documents 08:05 Modeling Documents with Milvus StructArray (Schema & Layout) 08:55 Dataset Walkthrough: WikiHow & Cosmopedia 11:04 Passage Matching & Fast Contextual Lookups (Neighboring Chunks) 16:42 In-Array Filtering & Element-Level Quantifiers 19:28 Server-Side Grouping & Deduplication by Document 22:30 Migration Guide: Reusing Chunks Without Re-Embedding 24:12 Trade-offs: When a Flat Schema Remains the Better Choice 27:00 Live Q&A: Ingesting Complex PDFs & Multi-Level Hierarchies 32:20 Live Q&A: Handling Technical Diagrams & Multimodal Data 37:10 Live Q&A: Cross-Document Linkages & Legal Citation Modeling 40:30 Live Q&A: Scripting the Flat-to-StructArray Migration 42:18 Live Q&A: Varying Document Lengths & Embedding Dimensionality 45:00 Live Q&A: Two-Pass Search (Entity First vs. Chunk First) 47:58 Live Q&A: Chunk Sizing & Best Practices for Dense Texts 54:40 Wrap-Up & $100 Free Zilliz Cloud Credits 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