Constructing Insights: Leveraging Large Language Models for Information Retrieval in Unstructured Document Collections with ConSight

Recent technological advances have led to an increase in the volume
of available data that knowledge workers use in their daily routines. Current LLM-
based systems face challenges in providing accurate and relevant information in an
efficient manner. At the same time, because these large collections of documents
are unknown to the user, many knowledge workers struggle to process the data,
even when collaborating with an LLM-based system. To address this challenge,
we develop a prototype "ConSight" following a Design Science Research method-
ology. Based on insights from the construction industry, our system integrates
information retrieval with multi-modal data support to enhance information access
and contextual understanding. Our contribution to the IS domain is twofold: (1)
we provide insights into the interplay between LLM-based retrieval systems and
human information processing in real-world scenarios, and (2) we take the first
steps to derive design knowledge for LLM-based knowledge work support.