Repository navigation
This is the first stable release of a robust Retrieval-Augmented Generation (RAG) system designed to "chat with your documents" using strictly local AI processing for maximum privacy and zero cost.
Unlike simple API wrappers, DocuMind implements a full Asynchronous ETL Pipeline capable of ingesting large PDF manuals without browser timeouts, powered by a modern Java 21 + React architecture.
🌟 Key Features
🧠 Intelligent RAG Engine
-
Zero Hallucinations: Engineered with a strict prompting strategy that forces the model to reply [[NO_INFO_FOUND]] if the answer isn't in the text.
-
Local Inference: Runs 100% locally using Ollama (Phi-3) and Nomic Embed Text, ensuring sensitive data never leaves your infrastructure.
-
Smart Context: Uses Top-K: 6 retrieval and optimized token chunking for precise answers.
⚡ Async Architecture & UX
-
"Fire-and-Forget" Ingestion: Uploads are handled asynchronously (202 Accepted). The server processes cleaning, chunking, and embedding in the background.
-
Real-Time Progress Tracking: (New) The UI features a dedicated progress bar component that polls the backend to show the exact processing status (0-100%) for large files, providing immediate visual feedback.