Retrieval-based chatbot for scholarly publications

Your challenge is to develop a retrieval-based chatbot designed to engage users in meaningful, interactive discussions about a given scholarly publication. The chatbot should be able to provide concise summaries and highlight key findings, while allowing users to ask follow-up questions about aspects such as the methodologies used, the main conclusions, or any potential limitations of the study. The primary objective is to ensure the chatbot can accurately interpret and extract insights from complex academic texts, delivering clear and reliable responses.

All publications for this challenge are open access, with no license restrictions, allowing participants to utilize OpenAI API endpoints (or similar commercial services). To support your development, an OpenAI API key with credits will be provided. Additionally, participants will receive a dataset containing several scholarly publications in JSON format, with structured sections such as “Introduction,” “Results,” “Methods,” “Conclusions,” and “References.”

The ideal solution will include retrieval-based safeguards, such as providing references and source links to avoid conversational derailment or "hallucination." This challenge offers an exciting opportunity to push the boundaries of NLP and AI in a practical, real-world context, making academic research more accessible and interactive for scholars, students, and the general public.