Agents Need More than Vector Search
summary
LanceDB presents a multimodal lakehouse that combines vector and full‑text indices, persisted to disk or object stores, and treats images/audio/video as first‑class blob columns. It claims 100–10,000x faster random access than Parquet for search, supports zero‑cost column evolution (backfills write only the new 1GB rather than copying 100GB), and powers a Netflix demo searching petabytes of video with results in under a second. Examples include hybrid retrieval, SQL via DuckDB, and pipelines built with frameworks like DSPy for multi‑hop and tool‑calling agents.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:34Agents in 2025
- 7:38Why Lance DB exists
- 11:26Lance DB overview
- 13:51Key benefits of Lance DB
- 19:51Agents and retrieval layers
- 23:21Evaluating agent pipelines
- 27:13Lance DB adoption and future
related talks
28mLanceDB: lightweight billion-scale vector search for multimodal AI
31mTransforming Multimodal Data Management with LanceDB-Ray
28mBeyond Parquet and ORC: Upgrading Data Infrastructure for Multi-modal AI with Lance Col...
27mUnified AI & Relational Power: Scalable Distributed Vector Search with
05mBringing AI to DuckDB with Lance columnar format for multi-modal AI
45mNever Send a Human to do an Agent's Search
16mThe Post-Modern Data Stack for AI: Open Source Foundations for a Multimodal La...
13mRAG Without the Hassle: Building AI-Powered Applications