Analyzing the Applications of Large Language Models in the Data Science Lifecycle

open_access
Open Access
Author:
Chintakunta, Sai Sanjna
Graduate Program:
Data Analytics
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
March 23, 2026
Committee Members:
Nathalia Moraes Do Nascimento, Thesis Advisor/Co-Advisor
Dusan Ramljak, Committee Member
Raghu Sangwan, Program Head/Chair
Everton Tavares Guimaraes, Thesis Advisor/Co-Advisor
Chengfei Wang, Committee Member
Keywords:
Large Language Models
Data Science
Multi-Agentic Framework
Empirical Evaluation
Abstract:
Large Language Models (LLMs) have shown great potential in performing a wide array of tasks, chief amongst them being solving advanced analytical problems. Their role in data science has attracted considerable attention, as LLMs demonstrate the potential to automate and enhance multiple stages of the data science lifecycle. Data science is the process of extracting knowledge and insights from both structured and unstructured data through analysis and algorithms. Their impact on the various tasks in data science has been explored by previous research. But the research remains fragmented, and does not consider the vast and varied nature of the types of problems in data science. The central focus of this thesis is to study the impact of LLMs on the data science lifecycle. This thesis holistically analyzes the influence LLMs have had on the Data Science Lifecycle. In order to ensure a well-rounded assessment, the thesis follows three major studies, each integral in uncovering important information. The studies are as follows: a Systematic Mapping Study to understand the current landscape, a Controlled Experiment to empirically evaluate the code generation quality of LLMs, and a Framework Design and Implementation of CodeViz, which is a system to automate data science tasks that serves as a foundation for building a more comprehensive solution.