ragent

Why we included this project

For Java teams building document-grounded question answering, this is a rare find: a full production-oriented Agentic RAG implementation rather than a thin demo wrapper. The pipeline runs from document ingestion to answered questions, with hybrid retrieval that pulls from vector, keyword, knowledge-graph, and web sources before reranking and fusing the results. What separates it from the usual tutorial code is the engineering around real-world failures: intent recognition and question rewriting, conversation memory backed by persisted summaries to control token costs, MCP tool discovery, and distributed concurrency control so a traffic spike doesn't overwhelm the model service. An admin console, answer tracing, user feedback, and a configurable ingestion pipeline round out the system. Developers who want to see how production RAG handles pitfalls, or teams hunting for a serious starting point for enterprise deployments, will find genuinely battle-tested patterns here.