RAG Template
A reference implementation for building a retrieval-augmented-generation chatbot and document management system, with a FastAPI backend, a document extraction pipeline and a choice of language model providers.
This is STACKIT's RAG Template, a working example of how to build an AI chatbot and document management system using retrieval-augmented generation. It is meant as a starting point rather than a finished product, built around a FastAPI backend, a container-based development setup and a helm chart for deployment.
It ingests PDFs, Office documents, spreadsheets, Markdown, e-books, HTML, XML, CSV and text files, and images, falling back between several extraction libraries, and it can also pull in Confluence spaces and sitemaps. Separate services handle the RAG backend, an admin backend for managing documents and Confluence spaces, document extraction, an MCP server exposing the RAG system, and a chat and admin frontend.
It supports multiple language model and embedding providers, including STACKIT and Ollama, stores documents in S3-compatible object storage, and integrates with Langfuse for monitoring.