Project collaborators
Shanan E. Peters, Geoscience
Shivaram Venkataraman, Ian Ross, Miron Livny, and Yuxiao Qu, Computer Sciences
Jason Lo, Data Science Institute
Project start and end dates
3/2023-5/2024
Project summary
Ask-xDD, funded by DARPA and USGS and developed in collaboration with the CHTC xdd team, offers an easy-to-use chat interface for accessing academic information. It helps bridge the gap between complex academic content and users, including laypeople and researchers. The project addresses issues like copyright restrictions and technical jargon, making scholarly articles more accessible. This is important in an era where misinformation spreads easily on social media.
Our current prototype, available at http://cosmos0001.chtc.wisc.edu:8501/, currently covers topics such as geoscience, climate change, and COVID-19. Ask-xDD uses retrieval augmented generation technology to provide accurate and relevant responses. It connects users to a vast database of scholarly articles at xdd.wisc.edu (more than 17M articles with full text), enabling informed, real-time answers to complex queries.
Our system integrates Weaviate and ElasticSearch to enable retrieval-augmented generation from a database of 17 million full-text journal articles. Leveraging Facebook’s Dense Passage Retriever and OpenAI’s ChatGPT for reasoning via ReAct, the system is accessible through a FastAPI endpoint, ensuring easy integration. Additionally, we provide an interactive Streamlit demo. For further information and access to the source code, visit https://github.com/UW-Madison-DSI/ask-xDD.