Offline vector search with SQLite and EmbeddingGemma
summary
Builds an offline browser RAG by encoding documents with EmbeddingGemma (38M parameters, 768‑dim), storing float32 vectors in SQLite VEC compiled to WASM, and syncing optional backups via Firestore vector support. Uses transformers.js/Onnx runtimes to run encoders locally, vectorizes a full Unicode emoji dataset and performs per‑keystroke queries, claims browser-side similarity search at massive scale (millions of comparisons per second), and demo shows it runs fully offline once the model is downloaded.
AI-written summary · about
chapters (9)
- 0:00Introduction to vectors
- 0:40Client vs server vectors
- 1:55Hybrid vector approach
- 3:10EmbeddingGemma overview
- 4:12Transformers.js integration
- 5:14Firebase vector storage
- 6:17SQLite vector search
- 8:05SQL query implementation
- 8:43Emoji search demo
related talks
01h 35mTutorial: Build AI-Supercharged RAG Apps with a Vector Database
50mA fun and absurd introduction to Vector Databases
48mA fun and absurd introduction to Vector Databases
46mA Fun & Absurd Introduction to Vector Databases
17mThink Fast: OnDevice Vector Features
48mA Fun & Absurd Introduction to Vector Databases
01h 23mVector Databases Demystified: Discover how they work
47mA fun and absurd introduction to Vector Databases