Retrieval-Augmented Generation usually means a vector database, a server, and a network call. None of that exists on a phone — and for offline-first apps or anything touching private user data, none of it should.
This talk is about building RAG entirely on-device: embedding text locally, storing and querying vectors without a server, and keeping the index fresh without re-downloading your whole knowledge base every time content changes. We'll cover the real constraints — model size vs. APK budget, query latency without a backend to do the heavy lifting, and the storage trade-offs that actually work on a mid-range phone, not just your dev device.
No cloud fallback, no asterisks. Just what it takes to make retrieval-augmented answers work entirely in your user's pocket.