Think Fast: OnDevice Vector Features
summary
Claims that React Native apps can run on-device vector features and do fast private searches offline, demonstrated by a working app that seeds a local SQL DB, runs image segmentation and embedding generation, and returns recommendations in real time. Uses Apple Vision framework for 768‑dimensional embeddings (Android 1,280 dims), Google MLKit style salience preprocessing, and an approximate vector distance SQL function. Live demo captures photos, stores metadata and images, and successfully matches items on a device with no network.
AI-written summary · about
chapters (13)
- 0:00Welcome
- 1:12What this talk is really about
- 2:56Flat data search limits
- 3:59Vectors in AI
- 5:44Vector search uses
- 7:29Grocery store example
- 8:31Product metadata with vectors
- 9:57Context improves search
- 10:59Image preprocessing
- 12:01Demo setup
- 13:50Live image search
- 14:51More use cases
- 15:34Daily life apps
related talks
07mMachine Learning on the Edge
10mMachine Learning And React Native
22mReact Native Local First Semantic Search
23mChallenges in Building React Native Applications with On-Device AI
10mOffline vector search with SQLite and EmbeddingGemma
23m2025: On‑Device AI Goes Mainstream on Android
26mBuild your Machine Learning Model on Edge with React Native
47mBoost Your Mobile Apps with Machine Learning and NativeScript