Skip to main content
Large Language Models

The Master’s in Applied Data Science Winter Capstone Project represents the innovative skills of students in addressing real-world industry challenges through data-driven methodologies. On March 2, MS in Applied Data Science students, faculty, alumni, industry partners, and judges came together as nine teams presented their capstone projects that spanned numerous industries, data types, and methodologies.

At the Winter Showcase two teams were selected by the judges for special recognition. Learn more about the winning projects below:

Deloitte: Using LLM to Build Industry Personas

Students: Davi Aragao, Nathan Deweerdt, Mayur Kumar, Sanket Mayekar

Advisor: Don Patchell

Davi Aragao, Nathan Deweerdt, Mayur Kumar, and Sanket Mayekar crafted a solution for Deloitte that focuses on creating industry specific large language model (LLM) personas to enhance Deloitte’s ability to tailor strategies, communications, and products/services for effective engagement with clients. The team said, “our financial analysis indicates that our solution could save up to $1.3 million in cost savings, with more added as it scales.”

The team expressed confidence in delivering a strong prototype, comprised of a comprehensive data pipeline, a user-tailored persona and prompt generator that can be customizable based on the user’s needs, front-end web application, and a thorough evaluation process for assessing the system’s performance. The initial testing results indicate that the prototype shows great promise.

Paula Payton, who was the team sponsor and head of Data Science and Survey Advisory at Deloitte said, “this team was so resourceful in terms of identifying data sources, and you certainty proved your mettle in terms of thinking very creatively about YouTube videos to find the voice of C-suite leaders.”

Circana: CPG Price Optimization

Students: Jupiter Angulo, Josefina Bollini, Ben Cox, Ezra Kim

Advisor: Anil Chaturvedi

For their project Jupiter Angulo, Josefina Bollini, Ben Cox, and Ezra Kim worked with Circana to pioneer a framework aimed at understanding consumer behavior and willingness to pay. “The implication for manufacturers, product developers, and retailers is that they need to figure out ways to more effectively reach the consumer where they are and figure out what incentivizes consumers to purchase things,” the team said.

Current methodologies used to achieve this task are considered old-fashioned, time consuming, and dependent on consumer surveys. The team noted that they developed a framework that is generalizable, scalable, and uses non-survey driven methods to understand the consumer preferences that Circana wanted to do a deeper dive on. The proposed methodology consists of utilizing a generalized utility model to calculate the value of attributes across multiple categories along with price elasticities using point of sale data and item attributes.

Honorable Mentions

Argonne: Real-time Object Detection for Avian-solar Interactions

Students: Aaron Chan, Aashish Singh, Alice Wang, Weiting Ye

Advisor: Igor Yakushin

Independent Research: Leveraging Retrieval-Augmented Generation

Students: Nathan Pak, Tyler Roth, Isaac Silver

Advisor: Ashish Pujari