WIP: First-year Student Support System: A Multi-agentic AI Approach

Rui Li

Download Paper | Permalink

Conference: FYEE 2025 Conference
Location: University of Maryland - College Park, Maryland
Publication Date: July 27, 2025
Start Date: July 27, 2025
End Date: July 29, 2025
Conference Session: WIP I
Tagged Topic: FYEE 2025
Page Count: 3
DOI: 10.18260/1-2--55279
Permanent URL: https://peer.asee.org/55279
Download Count: 6

Paper Authors

biography

Rui Li New York University

visit author page

Dr. Li earned his master's degree in Chemical Engineering in 2009 from the Imperial College of London and his doctoral degree in 2020 from the University of Georgia, College of Engineering.

visit author page

Download Paper | Permalink

Abstract

This research presents an innovative AI-powered educational support system designed to address diverse student needs through a hierarchical expert framework. First-year engineering students frequently face considerable hurdles as they go from high school to university. These problems can be academic, emotional, social, or psychological. Implementing a support system that is targeted to their specific requirements is critical for improving their overall well-being and academic performance.

The system employs a sophisticated two-tier semantic routing architecture that efficiently directs student queries to specialized AI experts across academic and wellness domains. The core of our system is a broker chatbot that serves as the initial point of contact, analyzing incoming queries using centroid vector representations of expert domains. This broker routes inquiries to one of two major groups—course-related or health-related—before connecting users to one of eight specialized experts (five academic experts and three health experts). Each expert maintains its own vector database collection that serves as domain-specific knowledge, significantly reducing hallucinations and improving response accuracy.

Our approach implements several technical innovations to enhance educational support. The hierarchical semantic router calculates centroid vectors from document embeddings to represent each expert’s domain knowledge, enabling automatic categorization of queries. This method eliminates the need for manually crafted utterances, instead deriving semantic understanding directly from the experts’ knowledge bases. The system enables multiple simultaneous users in a group chat environment, facilitating collaborative learning within student project team, and project information sharing. It also maintains conversation history to provide contextually relevant responses and builds user profiles to deliver personalized assistance. Security features include single sign-on authentication, VPN access requirements, and deployment on a Kubernetes cluster within network.

Performance evaluations demonstrate that our approach significantly improves response accuracy and reduces response time compared to traditional chatbot architectures. This research contributes to educational technology by demonstrating how specialized AI experts can provide comprehensive student support and assist course instructors to run the class. The system’s architecture balances computational efficiency with routing accuracy while maintaining an intuitive interface for both students and educators. A small-scale pilot study will start implementing in the fall of 2025.

Citation
Format

Li, R. (2025, July), WIP: First-year Student Support System: A Multi-agentic AI Approach Paper presented at FYEE 2025 Conference, University of Maryland - College Park, Maryland. 10.18260/1-2--55279

TY - CPAPER
AB - This research presents an innovative AI-powered educational support system designed to address diverse student needs through a hierarchical expert framework. First-year engineering students frequently face considerable hurdles as they go from high school to university. These problems can be academic, emotional, social, or psychological. Implementing a support system that is targeted to their specific requirements is critical for improving their overall well-being and academic performance.

The system employs a sophisticated two-tier semantic routing architecture that efficiently directs student queries to specialized AI experts across academic and wellness domains. The core of our system is a broker chatbot that serves as the initial point of contact, analyzing incoming queries using centroid vector representations of expert domains. This broker routes inquiries to one of two major groups—course-related or health-related—before connecting users to one of eight specialized experts (five academic experts and three health experts). Each expert maintains its own vector database collection that serves as domain-specific knowledge, significantly reducing hallucinations and improving response accuracy.

Our approach implements several technical innovations to enhance educational support. The hierarchical semantic router calculates centroid vectors from document embeddings to represent each expert’s domain knowledge, enabling automatic categorization of queries. This method eliminates the need for manually crafted utterances, instead deriving semantic understanding directly from the experts’ knowledge bases. The system enables multiple simultaneous users in a group chat environment, facilitating collaborative learning within student project team, and project information sharing. It also maintains conversation history to provide contextually relevant responses and builds user profiles to deliver personalized assistance. Security features include single sign-on authentication, VPN access requirements, and deployment on a Kubernetes cluster within network.

Performance evaluations demonstrate that our approach significantly improves response accuracy and reduces response time compared to traditional chatbot architectures. This research contributes to educational technology by demonstrating how specialized AI experts can provide comprehensive student support and assist course instructors to run the class. The system’s architecture balances computational efficiency with routing accuracy while maintaining an intuitive interface for both students and educators. A small-scale pilot study will start implementing in the fall of 2025.

AU - Rui Li
CY - University of Maryland - College Park, Maryland
DA - 2025/07/27
PB - ASEE Conferences
TI - WIP: First-year Student Support System: A Multi-agentic AI Approach
UR - https://peer.asee.org/55279
DO - 10.18260/1-2--55279
ER -