Question answering about Research Data Management

Presentation and code of team TLDR, which participated in Challenge 3: Question answering about Research Data Management of the Dataxplorers Hackathon 2023. The challenge was provided by members of the NFDI4Earth consortium for Earth System Sciences. The software attached employs Large Language Models in conjunction with custom embeddings to answer the questions of users correctly and in a concise and understandable way. The application also allows for the creation of new embeddings based on existing websites, as well as some basic functionality to perform preliminary checks on .csv-files against data curation guidelines. 

It must be noted that the Large Language Models employed are made available through the blablador API provided by Helmholtz AI workgroup at FZ Jülich. Thus, the application requires a valid access token to the blablador API which is as of now only available for members of the Helmholtz community. Since the API is closely related to the API openAI employs, customizing the application to work with openAI and/or other providers should be possible.