Towards the improvement of project team performance based on large language models

The subject of the study is a method for identifying poor quality project sprint task descriptions to improve team performance and reduce project risks. The purpose of the study is to improve the quality of textual descriptions of sprint tasks in tracking systems by implementing models for identifying and improving potentially poor task descriptions. Research Questions: 1. Can poor quality project sprint task descriptions be identified using clustering? 2. How to utilize the power of large language models (LLMs) to identify and improve textual descriptions of tasks? Objectives: to analyze research on approaches to improving descriptions using clustering and visualization techniques for project tasks, to collect and prepare textual descriptions of sprint tasks, to identify potentially poor task descriptions based on clustering their vector representations, to study the effect of prompts on obtaining vector representations of tasks, to improve task descriptions using LLMs, and to develop a technique for improving project team effectiveness based on LLMs. Methods of vector representation of texts, methods of dimensionality reduction of PCA and t-SNE data space, methods of agglomerative clustering, methods of prompting were used. The following results were obtained. An approach to improving the performance of the project team based on the use of LLM was proposed. Answering the first research question, it was found that there are no linguistic features affecting the perception of textual descriptions of project sprint tasks. In response to the second research question, a model for identifying potentially poor task descriptions is proposed to reduce project risks associated with misunderstanding of task context. Conclusions. The results suggest that project sprint task descriptions can be improved by using large-scale language models for project team understanding. Future research recommends using project source documentation and project context as a vector repository and source of context for LLM. The next step is to integrate the LLM into the project task tracking system.