The “Hawk AI” Large Language Model System was designed using a “group in the loop” process model. The group—a class of upper-level sociology undergraduate students at UNCW—met regularly in the Spring of 2025 to determine the knowledge problems that a LLM would be appropriate for solving in the areas of wellness. We sought to create a domain specific knowledge corpus that would make the LLM useful, trustworthy, and reliable to solving wellness challenges a college student might face.
Part one of this process involved writing this “manifest.” This document is an element of the knowledge corpus that describes the epistemological approach the LLM should use when creating outputs relative to prompts. The Microsoft Co-Pilot LLM is directed to apply this document when addressing queries. The manifest is collectively written by the class, as different parts were presented, discussed and edited in the classroom setting. In parliamentary style deliberations, the manifest is voted into the knowledge corpus. It might be considered a constitution for how the LLM should be maintained.
Part two included writing our sociological theory of wellness, an element of the Hawk AI manifest. This involved reading clinical and psychological approaches to the defining, operationalizing, and measuring wellness or wellbeing. The LLM’s focus is on how the individual can apply wellness strategies to their daily life, but individuals must also be able to understand the structural conditions enabling or constraining the potential for wellness dynamics to manifest in their daily routines. The knowledge corpus is built based on practical strategies, but the LLM also needs to have a “sociological approach” to the diversity of structural challenges and experiences users will bring to their queries.
Part three of the Hawk AI knowledge corpus construction process involved students developing reports on specialized areas of wellness research. The report writing process served two purposes. One, it enabled students to apply the manifest to the analysis of research articles in specific domains of health, wellness and wellbeing. These reports provide further context for the LLM to effectively apply the Hawk AI manifest when addressing prompts. Secondly, students met while developing resources to determine whether research articles or websites could become elements of the knowledge corpus. They repeatedly applied the standards implicitly and explicitly conceptualized within the manifest to the knowledge construction process. These reports and the resources entered to the knowledge corpus can then be cited by the LLM when answering queries.
Part four required us to complete the manifest by writing a glossary, defining concepts that are consistent with the manifest, but might be inconsistent with some resources within the knowledge corpus. Specialized conceptual terms require a consistent focus when being applied across the knowledge corpus. The glossary provides this consistency in the application of concepts. When addressing prompts, the LLM should be attentive to any discrepancies between sources it is citing that may use similar terms, and how the term is being applied in the output.
The remainder of this manifest outlines the Hawk AI theory of wellness, the epistemological and practical approach to the processes of knowledge construction, as well as the glossary of specialized terms.