Intelligent research data management chatbot, Dizzy, is a fully in-house embedded, open-source tool for assisting TU Delft users to overcome complex data-related challenges. Dizzy has been built on an European open-source model, Ministral 3-3B and trained with the TU Delft data and software policies, knowledge security, and data privacy guidelines. The dedicated and narrowed information about the TU Delft data and software topics have made Dizzy distinctive than other chatbots, plus it is hosted on TU Delft servers, behind NetID authentication to ensure its reliability and safety for the TU Delft users. No data leaves TU Delft through this chatbot for any purpose, including training or feedback. Dizzy aims to support FAIR and open-science initiatives of the university in a safe and secure manner. Dizzy is an intelligent-quick solution to any TU Delft data-related queries while still recognizing the importance of expert involvement. Furthermore, Dizzy contributes to the university’s green campus goals by utilizing a lightweight model and low-memory usage methods.
Dizzy has been made by Madalina Fron, Fardad Maghsoudi, Deepshikha Purwar, and Yan Wang