PhD Candidate · Computer Science

Tahreem Yasir

I study how large language models can support learning in structured domains — especially intelligent tutoring, step-level feedback, and the design of student-facing AI tools in computing education.

North Carolina State University · Raleigh, NC

Google Scholar LinkedIn GitHub Email

About

I am a PhD student in Computer Science at North Carolina State University, where I work on generative AI for education — evaluating LLM-based tutoring agents, multi-agent feedback pipelines, and how teachers integrate AI tools in the classroom.

Before graduate school, I was a Developer Advocate at Educative, creating courses on cloud infrastructure and data analysis, and an Assistant Professor at the National University of Computer and Emerging Sciences (FAST-NU) and the University of Management and Technology, where I taught software engineering and HCI and published work on machine learning and accessible technology.

My research interests include intelligent tutoring systems, LLM evaluation in symbolic reasoning domains, computing education, and human-centered AI design.

Recent Updates

  1. Paper accepted at BEA @ ACL 2026

    Our study of seven LLM tutoring agents examines where feedback systems succeed—and where they miss valid or incorrect student reasoning.

  2. New preprint on multi-agent verification

    We find that verification can improve weak upstream feedback while degrading feedback that is already reliable.

  3. Systematic review released

    Our review synthesizes 52 studies of student-facing LLM tools in undergraduate computing education.

Research Themes

LLM Tutoring & Feedback

Evaluating whether LLM agents can distinguish optimal, valid-alternative, and incorrect student reasoning — and when multi-agent verification helps or hurts tutoring quality.

Computing Education

Systematic reviews and teacher-facing frameworks for integrating student-facing LLM tools in undergraduate and secondary computing classrooms.

Knowledge-Grounded AI

Hybrid architectures that combine knowledge graphs and expert solution paths with LLMs for more reliable, pedagogically actionable feedback.

Publications

First-Author Papers

Listed in reverse chronological order.

Figure 1: LLM tutoring evaluation pipeline Figure 1 · Tutoring evaluation pipeline
First Author Accepted BEA @ ACL 2026

Confirming Correct, Missing the Rest: LLM Tutoring Agents Struggle Where Feedback Matters Most

Tahreem Yasir, Wenbo Li, Sam Gilson, Sutapa Dey Tithi, Xiaoyi Tian, Tiffany Barnes

Accepted at the 21st Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2026), co-located with ACL 2026 · arXiv:2605.16207

We benchmark seven LLM feedback agents in propositional logic using knowledge-graph-derived ground truth across 10,836 solution–feedback pairs. Models excel on optimal steps but systematically over-reject valid alternatives and over-validate incorrect solutions — precisely where adaptive tutoring matters most.

arXiv Code
Figure 1: Multi-agent tutoring with KG-grounded hints Figure 1 · Multi-agent tutoring pipeline
First Author Preprint

When Verification Hurts: Asymmetric Effects of Multi-Agent Feedback in Logic Proof Tutoring

Tahreem Yasir, Sutapa Dey Tithi, Benyamin Tabarsi, Dmitri Droujkov, Sam Gilson, Yasitha Rajapaksha, Xiaoyi Tian, DongKuan Xu, Tiffany Barnes, Arun Ramesh

arXiv preprint arXiv:2603.27076, 2026

We introduce a knowledge-graph–grounded benchmark of 516 proof states and evaluate role-specialized tutoring pipelines. Verification improves outcomes when upstream feedback is error-prone, but degrades performance when feedback is already reliable — revealing a shared complexity ceiling for step-level tutoring.

arXiv

Co-Authored Papers

Collaborative work; first author listed first.

Figure 1: PRISMA flow diagram for systematic literature review Figure 1 · PRISMA review flow
Co-Author Preprint

Herald of Advancement, Disruption, or Both: A Systematic Literature Review on Student-Facing LLM Tools in Undergraduate Computing Education

Benyamin Tabarsi, Tahreem Yasir, Heidi Reichert, Xiaoyi Tian, Shiva Gadireddy, Clara DiMarco, Daniel Briceno, Tiffany Barnes

TechRxiv preprint, 2025 · Under review at Computers and Education: Artificial Intelligence

A PRISMA-guided review of 52 papers on purpose-built, student-facing LLM tools in undergraduate computing education — covering pedagogical roles, technical design, evaluation methods, and impacts on learners.

Zenodo
Figure 1: Framework for LLM integration in secondary computing classrooms Figure 1 · Six-phase integration framework
Co-Author Conference

A Framework for LLM Integration in Secondary Education: Insights from Computing Teachers

Heidi Reichert, Tahreem Yasir, Malvika Satyavolu, Tiffany Barnes

International Conference on Artificial Intelligence in Education (AIED), 2026

Through interviews and surveys with 11 secondary computing teachers, we develop an empirically grounded framework of 27 implementation steps across six phases — Prepare, Adapt Curriculum, Lead, Assess, Synthesize, and Share — for integrating LLM tools in classrooms.

Contact

I am happy to connect about research collaborations, tutoring systems, or computing education. Reach me at tahreemyasirr@gmail.com.

Email Google Scholar LinkedIn GitHub