Leveraging open-source technologies for a retrieval-augmented generation chatbot in Greek newsrooms

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This paper presents a Retrieval-Augmented Generation chatbot application tailored for Greek newsrooms. The system leverages the open-source Greek LLM KriKri along with open-source embedding models, emphasizing newsroom autonomy by enabling the use of AI without reliance on external providers. The chatbot is designed to assist journalists in their investigations by delivering fact-based answers accompanied by cited sources from reputable Greek news outlets. The app can be run locally, making it an accessible and privacy-preserving AI solution for newsrooms. The dataset consists of articles sourced from four different Greek news websites. The system’s architecture is described, encompassing the construction of a vector database for source retrieval, the integration of an instruction-tuned model, and the implementation of a chatbot interface that displays linked sources. An initial evaluation with senior Journalism students indicated positive perceived usefulness, ease of use, satisfaction, and willingness to adopt the tool, highlighting the importance of independent, localized, transparent, and trustworthy AI solutions for journalism.