Advancing AI for Education through LLMs, BERT, and Student Behavior Modeling

About Me

I am a Ph.D. student in Computer Science at the University of Memphis and a Graduate Assistant under Dr. Deepak Venugopal. My research investigates the use of large language models (LLMs) to model and simulate student learning behaviors, integrating BERT and Deep Knowledge Tracing within intelligent tutoring systems in collaboration with Carnegie Learning. To learn more about my research, please visit ASTRA

My work has been accepted at LAK (Learning Analytics and Knowledge Conference) and is aimed at venues such as AIED (Artificial Intelligence in Education) and Learning@Scale.

WORK EXPERIENCE

Graduate Assistant (Research & Teaching)

University of Memphis, Memphis, TN
Jan 2024 – Present

Research Responsibilities:

Conducting research in Artificial Intelligence and Education, focusing on student modeling, strategy prediction, and intelligent tutoring systems in collaboration with Carnegie Learning.
Developing Large Language Model (LLM)–based frameworks for predicting student learning strategies, hint behaviors, and mastery trajectories.
Leading experiments integrating BERT fine-tuning and Deep Knowledge Tracing (DKT) to model problem-solving strategies in MATHia.
Authored “Understanding and Modeling Math Strategy Use in Intelligent Tutoring Systems” (submitted to LAK 2026).

Teaching Responsibilities:

Assisting in undergraduate and graduate courses on Artificial Intelligence, Machine Learning, and Database Systems, supporting around 65 students each semester.
Mentoring students on course projects, conceptual understanding, and applied problem-solving.
Supporting course logistics, grading, while integrating AI examples into assignments and tutorials.

Software Engineer
TechnoNext Ltd. (US-Bangla Group), Dhaka, Bangladesh
May 2022 – Nov 2023

Designed and deployed enterprise-level software solutions integrating data analytics and automation.
Developed machine-learning modules for performance prediction and process optimization.
Collaborated with cross-functional teams to ensure efficient backend integration and user-centered design.

Software Engineer
Sheba Platform Limited, Banani, Dhaka, Bangladesh
Jan 2021 – May 2022

Developed information systems by defining scope, gathering requirements, and delivering scalable solutions.
Built and maintained React-based applications in collaboration with system architects.
Implemented Agile methodologies (daily stand-ups, sprint planning) to improve team coordination and delivery timelines.

EDUCATION

Ph.D. in Computer Science
University of Memphis, Memphis, TN, USA
Jan 2024 – Expected May 2028

Research focus: Artificial Intelligence in Education, Student Modeling, and Large Language Models (LLMs).
Working under the supervision of Prof. Deepak Venugopal in collaboration with Carnegie Learning.
Ongoing project: Understanding and Modeling Math Strategy Use in Intelligent Tutoring Systems (submitted to LAK 2026).

M.Sc. in Computer Science
University of Memphis, Memphis, TN, USA
Jan 2024 – Dec 2025

Specialized in Machine Learning, Natural Language Processing, and Educational Data Mining.
Completed thesis research on BERT-based student behavior modeling and hint classification.

B.Sc. in Computer Science and Engineering
BRAC University, Dhaka, Bangladesh
2016 – 2020

Focused on Artificial Intelligence, Software Engineering, and Data Analytics.
Conducted undergraduate research on deep learning for image-based disease detection.

Publications

1. Understanding and Modeling Math Strategy Use in Intelligent Tutoring Systems (Paper)
Abisha Thapa Magar, Asad Uzzaman, Deepak Venugopal, Dr. Vasile Rus, Dr. Stephen Fancsali, Dr. Steve Ritter...
Learning Analytics and Knowledge Conference (LAK), 2026

This work investigates how students employ multiple mathematical strategies, such as Equivalent Ratios (ER) and Means & Extremes (ME), in the MATHia intelligent tutoring environment. We present a BERT-based model fine-tuned on student action sequences and Deep Knowledge Tracing (DKT) for predicting next-strategy use and hint behavior. The study offers empirical insights into student learning dynamics and contributes to the development of interpretable AI models for education.

Fig. 1: Results generated by Claude Sonnet 4 that show the results of analyzing the effect of the intervention on learning to apply the ER and ME strategies in the appropriate context. (a) shows the results for ER and (b) for ME.

Fig 2: Divided the students into 9 bins shown based on time-to-mastery, i.e., the number of problems they had to attempt before graduating from the ER(pre) and ME(pre) workspaces, respectively.

2.Automatic Ocular Disease Detection Scheme from Enhanced Fundus Images Based on Ensembling Deep CNN Networks (Paper)
Ishtiaque Ahmed Khan, Asad Uzzaman, Shaikh Anowarul Fattah
Proceedings of the 11th International Conference on Electrical and Computer Engineering (ICECE), 2020
Developed an ensemble deep learning framework for classifying ocular diseases from retinal fundus images, improving diagnostic performance across multiple pathological classes.

Fig. 3: Flow chart of the proposed method.

Table: Performances on the model of Raw and Enhanced Images

Fig.4: Sample images detected by the proposed method.

3.Parkinson’s Disease Detection Using fMRI Images Leveraging Transfer Learning on Convolutional Neural Networks (Paper)
Asad Uzzaman et al.
Proceedings of the International Conference on Machine Learning and Cybernetics (ICMLC), 2020
Proposed a CNN-based transfer learning approach to detect early signs of Parkinson’s disease using fMRI images, demonstrating improved generalization across limited medical datasets.