A new Private On-Device RAG System has been developed, allowing for private conversations with documents without needing an internet connection.
The system combines natural language understanding of LLMs with keyword-based limitations of traditional document search for contextually aware answers.
The system runs locally, consisting of five components including file ingestion, embedding, semantic query understanding, memoization, and local answer generation.
The project emphasizes the importance of personal privacy by showcasing that powerful AI can coexist with secure and private systems.