eGovFrame AI RAG Samples
Two side-by-side sample RAG (Retrieval-Augmented Generation) projects for Korea's eGovFrame standard framework, one built with Spring AI and Redis Stack and the other with LangChain4j and PostgreSQL/pgvector.
This repository provides two reference implementations of the same Retrieval-Augmented Generation question-answering system for eGovFrame, Korea's government standard framework, built with different technology stacks: one using Spring AI with a Redis Stack vector store, the other using LangChain4j with PostgreSQL and the PGVector extension.
Both samples target JDK 17 and Spring Boot, and run local models through Ollama (version 0.17.1 or later, since earlier versions have a memory-exposure vulnerability in GGUF model handling), with Docker used for the vector store. Tested embedding and language models include a Korean SBERT-style model, Google's EmbeddingGemma, and several open-weight LLMs such as GPT-OSS, HyperCLOVA X, EXAONE, Kanana, and Qwen3, run on both Windows and macOS hardware.
Setup requires installing Ollama and the chosen LLM, exporting an ONNX embedding model, and starting the Docker-based vector store; a Modelfile lets users customize model behavior for offline environments.