SpiritRAG is a domain‑focused question‑answering system that connects
state-of-the-art large language models with more than 7,500 United Nations
documents related to religion, spirituality, health, and education. It is
designed to support rigorous, document‑grounded knowledge acquisition in
settings where concepts are context‑dependent, contested, and historically
situated.
Rather than acting as a general‑purpose chatbot, SpiritRAG operates as a
research infrastructure: it retrieves relevant resolutions, reports, and
policy texts from a carefully curated corpus, and then uses retrieval‑augmented
generation to synthesise answers that remain anchored in those sources. This
makes it possible to explore how religion and spirituality are framed across
different UN bodies, time periods, and policy domains, while preserving
explicit links back to the underlying documents.
SpiritRAG is intended for scholars, students, and practitioners who need a
transparent and productive way to query large text collections without
losing sight of the original context. By combining semantic search,
citation‑style outputs, and domain boundaries tailored to R&S research, it
aims to facilitate careful reading, comparative analysis, and interdisciplinary
work at the intersection of religion, spirituality, global governance, and
public policy.
SpiritRAG is constrained to questions related to religion, spirituality, health and education. Off‑topic queries are filtered out, helping ensure that responses remain within the intended research domain.
The system combines semantic retrieval with large‑language models. For each query, relevant UN resolutions, policy texts and other curated sources are retrieved and used to generate answers, with references back to the original material.
SpiritRAG is developed within an academic environment, with a focus on transparency, productivity and careful domain boundaries. It is intended to support scholarly work rather than open‑ended general chat.
The online SpiritRAG Manual offers a more detailed overview of the project than this landing page. It introduces the research context, the data and engineering choices behind SpiritRAG, and the features currently available to users.
You can read the full manual on GitBook: SpiritRAG Manual .
SpiritRAG was presented at EMNLP 2025 (System Demonstrations Track) in Suzhou, China and 2025 ETH AI + X Summit.
If you use SpiritRAG in your work, please cite:
@inproceedings{gao2025spiritrag,
title={SpiritRAG: A Q\&A System for Religion and Spirituality in the United Nations Archive},
author={Gao, Yingqiang and Winiger, Fabian and Montjourides, Patrick and Shaitarova, Anastassia and Gu, Nianlong and Peng-Keller, Simon and Schneider, Gerold},
booktitle={Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations},
pages={26--41},
year={2025}
}
For questions, feedback, or collaboration inquiries related to SpiritRAG, please use the form below or contact the project team directly.
