✳flâneur — a map of the web's best reading

Lance Format and LanceDB: Columnar Storage for the Embedding Age | Andrea Bozzo | Blog

andreabozzo.github.io · 1,674 words · saved by 1 readers

Lance is a columnar format built for ML workloads — fast random access, native vector indexing, and zero-copy Arrow integration. This article walks through the format, LanceDB, and wiring it into a live NATS stream.

Columnar storage for the embedding age Table of Contents Lance Format and LanceDB: Columnar Storage for the Embedding Age # I have been spending time lately looking at storage formats from the perspective of ML infrastructure rather than analytics. Most of my day-to-day work sits in the Arrow/DataFusion ecosystem, so when I wanted to add a vector search layer to a small event streaming project, my first instinct was to ask: is there a format that speaks Arrow natively, handles embeddings without a separate vector database, and doesn't require me to run a separate server? Lance turned out to be

Explore this link on the map →

saved by

related reading