The Miktam Bible app has developed an AI study assistant that operates entirely on a user's phone, eliminating the need for cloud API wrappers. This assistant utilizes the SmolLM2-135M model, quantized to GGUF format and running via llama.rn, a React Native binding for llama.cpp. To ensure accuracy and prevent hallucination, the app employs a retrieval-augmented generation (RAG) approach, where user questions are first used to search an offline Bible database for relevant verses. These verses are then incorporated into the prompt for the on-device model to generate a grounded answer, with references linked back to the Bible text. Future updates aim to incorporate hybrid search, combining keyword and vector similarity, and the app is currently in closed testing on Google Play. AI
IMPACT Enables private, offline AI functionality on mobile devices by running models locally.
RANK_REASON This is a description of a specific application built using existing AI models and techniques, rather than a release of a new model or significant research.
- AI2TH
- android
- FTS5
- GGUF
- GitHub
- Google Play
- John 3:16
- llama.cpp
- llama.rn
- Miktam Bible
- MiniLM
- SmolLM2-135M
- SQLite-vec
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
