Vector Search on Android — Finding What You Mean, Not What You Type

Keyword search is great until the user searches "red shoes" and your database returns everything tagged "red" or "shoe" sorted by nothing. Vector search fixes that by finding things by meaning, not by string matching. And the best part is that it runs entirely on-device, reducing latency, preserving user data privacy and making it an offline capable solution.

This talk covers the three ingredients you need and nothing more:
1. An embedding model
2. A vector database
3. Chunking

We'll walk through concrete use cases: semantic search in notes taking app, textual lookups in images ("find products that look have red shoes"). Code snippets, and links to guides can be referenced, instead of slides full of math. We will also talk about some of the caveats, such as bundling with your artifact or downloading a model into your app storage and memory to perform the embedding work.