pith. sign in

arxiv: 2604.24807 · v1 · submitted 2026-04-27 · 💻 cs.CY · cs.AI· cs.MA

From Prototype to Classroom: An Intelligent Tutoring System for Quantum Education

Iizalaarab Elhaimeur , Nikos Chrisochoides This is my paper

Pith reviewed 2026-05-08 01:34 UTC · model grok-4.3

classification 💻 cs.CY cs.AIcs.MA
keywords intelligent tutoring systemsquantum computing educationmulti-agent LLM systemscurriculum analyticsclassroom deploymentquantum information scienceAI in education
0
0 comments X

The pith

A multi-agent tutoring system with specialized agents operates reliably in a real quantum course and reveals curriculum gaps to instructors.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Quantum computing courses face counterintuitive concepts, dense math, and limited qualified instructors. The paper refines an earlier LLM-agent prototype by adding agent specialization, a full curriculum, cloud infrastructure, and analytics. In a live course deployment, the changes eliminated earlier reliability failures, supported many students at low cost, and gave the instructor new visibility into where students were struggling. This addresses whether such systems can move from demonstration to practical classroom use.

Core claim

The authors present ITAS, built on a five-module curriculum, a Spoke-and-Wheel architecture of quantum-specialized agents, production cloud infrastructure, and a conversational analytics layer. Pilot use in a university quantum course shows the specialized agents handle the full range of tasks without the boundary failures seen in the unspecialized prototype, the infrastructure sustains classroom-scale use below textbook cost, and the analytics surface curriculum gaps the instructor could not otherwise detect.

What carries the argument

The Spoke-and-Wheel teaching architecture, in which a central coordinator routes queries to distinct quantum-specialized LLM agents each responsible for one tutoring function such as dynamic teaching or lesson planning.

If this is right

  • Specialized agents overcome the task-boundary reliability problems that general LLM prototypes encounter in technically demanding subjects.
  • Cloud infrastructure supports concurrent use by an entire class while remaining cheaper than printed materials.
  • Conversational analytics supply instructors with previously unavailable data on student difficulties and curriculum effectiveness.
  • The approach could expand quantum education access beyond institutions with dedicated faculty.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar specialization patterns could strengthen tutoring tools in other advanced technical fields facing expert shortages.
  • Aggregated analytics across multiple courses might support data-driven curriculum revisions at the program level.
  • Adding direct links to student assessment records could increase the precision of the identified gaps.
  • Low operating costs could make the system practical for smaller or less-resourced programs.

Load-bearing premise

The improvements observed in one course will appear in other courses with different students and instructors, and the analytics will correctly identify real curriculum gaps without separate validation.

What would settle it

Deploy the system in a second independent course and compare the analytics outputs against separate instructor reviews or student performance data on the same topics to check for agreement.

Figures

Figures reproduced from arXiv: 2604.24807 by Iizalaarab Elhaimeur, Nikos Chrisochoides.

Figure 1
Figure 1. Figure 1: The ITAS student interface. Three panels (video player, view at source ↗
Figure 2
Figure 2. Figure 2: The Spoke-and-Wheel teaching architecture. Three special view at source ↗
Figure 3
Figure 3. Figure 3: The analytics architecture. A single conversational agent view at source ↗
Figure 4
Figure 4. Figure 4: Estimated median end-to-end response latency vs. concurrent view at source ↗
Figure 5
Figure 5. Figure 5: Estimated per-student cost per semester compared to a typical view at source ↗
read the original abstract

Quantum computing instructors face a compounding problem: the concepts are counterintuitive, the mathematical formalism is dense, and qualified faculty are scarce outside a small number of well-resourced institutions. Our prior work introduced a knowledge-graph-augmented tutoring prototype with two specialized LLM agents: a Teaching Agent for dynamic interaction and a Lesson Planning Agent for lesson generation. Validated on simulated runs rather than in a real course, that prototype left open whether more aggressive agent specialization would be needed to handle the full range of quantum education tasks under real student load. This paper answers the three questions that the prototype could not answer. Can agent specialization solve the reliability problem in a domain as technically demanding as quantum information science? Can the system run in a real course, not a demonstration? Does the instructor gain actionable intelligence from the deployment? We present ITAS (Intelligent Teaching Assistant System), a multi-agent tutoring system built around four contributions: a five-module QIS curriculum grounded in Watrous's information-first framework, a Spoke-and-Wheel teaching architecture with quantum-specialized agents, a cloud infrastructure designed for production use and regulatory compliance, and a conversational analytics layer for instructors and content developers. Piloted in a quantum computing course at Old Dominion University, the system supports all three answers: deployment evidence is consistent with specialization addressing the task-boundary failures observed in the prototype, cloud infrastructure supports classroom-scale concurrency at sub-textbook cost, and the analytics agent surfaces curriculum gaps the instructor could not otherwise see.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces ITAS, a multi-agent intelligent tutoring system for quantum education featuring a Spoke-and-Wheel architecture with specialized LLM agents, a five-module curriculum based on Watrous's information-first framework, cloud infrastructure for production use, and a conversational analytics layer. Building on a prior prototype that was only simulated, it reports a pilot deployment in a quantum computing course at Old Dominion University and claims that the evidence is consistent with agent specialization resolving task-boundary reliability failures, that the system supports classroom-scale concurrency at sub-textbook cost, and that the analytics layer surfaces actionable curriculum gaps for instructors.

Significance. If the pilot results can be substantiated with quantitative metrics and appropriate controls, the work would be significant for AI-assisted education in technically demanding domains such as quantum information science, where faculty shortages are acute. It provides a concrete example of moving from simulated prototype to real classroom deployment, including infrastructure and analytics components that could serve as a template for similar systems in other STEM fields.

major comments (2)
  1. [Pilot results / Abstract] The central claim that 'deployment evidence is consistent with specialization addressing the task-boundary failures observed in the prototype' lacks any reported quantitative metrics (e.g., task failure rates, completion statistics, student performance deltas, or direct comparison to the non-specialized prototype). This makes it impossible to evaluate whether observed stability is attributable to the Spoke-and-Wheel architecture rather than the five-module curriculum, instructor scaffolding, or the specific student cohort.
  2. [Abstract and Conclusions] The manuscript asserts that benefits will generalize beyond the single-site pilot at Old Dominion University without providing controls, baseline comparisons, or discussion of limitations in student population or course context, which is load-bearing for the claim that the system 'supports all three answers' regarding reliability, scalability, and analytics utility.
minor comments (1)
  1. [System Architecture] The Spoke-and-Wheel architecture is referenced repeatedly but would benefit from an explicit diagram or formal definition early in the manuscript to clarify the roles of the specialized agents relative to the prior prototype.

Simulated Author's Rebuttal

2 responses · 1 unresolved

Thank you for the referee's constructive and detailed review. We address the major comments point by point below, indicating where revisions will be made to the manuscript.

read point-by-point responses

  1. Referee: [Pilot results / Abstract] The central claim that 'deployment evidence is consistent with specialization addressing the task-boundary failures observed in the prototype' lacks any reported quantitative metrics (e.g., task failure rates, completion statistics, student performance deltas, or direct comparison to the non-specialized prototype). This makes it impossible to evaluate whether observed stability is attributable to the Spoke-and-Wheel architecture rather than the five-module curriculum, instructor scaffolding, or the specific student cohort.

    Authors: We agree that the absence of quantitative metrics limits the strength of the causal attribution. The pilot was an observational deployment in a live course; system logs documented the absence of the specific task-boundary failures seen in the earlier simulated prototype, but no pre/post failure rates, completion statistics, or controlled comparison to a non-specialized version were collected. In the revised manuscript we will expand the deployment results section with additional qualitative examples from the logs and instructor notes, explicitly discuss alternative explanations (curriculum design, instructor scaffolding, and cohort characteristics), and revise the abstract language from 'consistent with' to 'suggestive of' while adding a limitations paragraph. We cannot add numerical metrics that were not gathered during the pilot. revision: partial

  2. Referee: [Abstract and Conclusions] The manuscript asserts that benefits will generalize beyond the single-site pilot at Old Dominion University without providing controls, baseline comparisons, or discussion of limitations in student population or course context, which is load-bearing for the claim that the system 'supports all three answers' regarding reliability, scalability, and analytics utility.

    Authors: The manuscript presents the study as a single-institution pilot and ties its claims to evidence observed in that setting. We acknowledge that the abstract and conclusions would benefit from stronger qualification. In revision we will (1) qualify the abstract to state that the three answers are supported by evidence from the Old Dominion University pilot, (2) add an explicit limitations subsection addressing the single-site context, small cohort size, lack of controls or baselines, and course-specific factors, and (3) adjust the phrasing of 'supports all three answers' to reflect the scope of the pilot data. These changes will make the generalizability caveats load-bearing rather than implicit. revision: yes

standing simulated objections not resolved
  • Absence of quantitative metrics (task failure rates, completion statistics, or controlled comparisons) from the original pilot data, which prevents direct substantiation or statistical evaluation of the specialization claim.

Circularity Check

1 steps flagged

Minor self-citation to prior prototype; central claims rest on independent pilot evidence

specific steps
  1. self citation load bearing [Abstract]
    "Our prior work introduced a knowledge-graph-augmented tutoring prototype with two specialized LLM agents: a Teaching Agent for dynamic interaction and a Lesson Planning Agent for lesson generation. Validated on simulated runs rather than in a real course, that prototype left open whether more aggressive agent specialization would be needed to handle the full range of quantum education tasks under real student load."

    The citation to the authors' own prior prototype supplies background motivation but does not bear the load of the central claims; those claims are supported by new deployment evidence from the current pilot rather than reducing to quantities or definitions established in the cited work.

full rationale

The paper describes a new multi-agent system (ITAS) and reports results from its pilot deployment in a quantum computing course at Old Dominion University. It references the authors' prior prototype paper only to establish the open questions (task-boundary failures under real load) that the current work addresses. The three answers claimed—specialization solving reliability issues, cloud infrastructure supporting classroom scale, and analytics surfacing curriculum gaps—are tied directly to observations from the new pilot rather than any mathematical reduction, parameter fit, or definitional equivalence to the prior work. No equations, fitted inputs presented as predictions, uniqueness theorems, or ansatzes appear in the text. This is a standard minor self-citation that does not render the derivation circular.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The claims rest primarily on domain assumptions about LLM agent reliability in technical education and the validity of pilot observations, with no fitted parameters or new physical entities postulated.

axioms (1)
  • domain assumption Specialized LLM agents can reliably handle the full range of quantum information science education tasks under real student load
    Invoked to justify the spoke-and-wheel architecture and to claim resolution of prototype task-boundary failures.
invented entities (1)
  • Spoke-and-Wheel teaching architecture no independent evidence
    purpose: To coordinate multiple quantum-specialized agents for dynamic tutoring
    New architectural pattern introduced in this paper to organize the multi-agent system.

pith-pipeline@v0.9.0 · 5573 in / 1417 out tokens · 61807 ms · 2026-05-08T01:34:47.981910+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    Preparing for the quantum revolution: What is the role of higher education?

    K. Fox, B. M. Zwickl, and H. J. Lewandowski, “Preparing for the quantum revolution: What is the role of higher education?” Physical Review Physics Education Research, vol. 16, p. 020131, 2020

    work page 2020

  2. [2]

    Disparities in access to U.S. quantum information education,

    J. C. Meyer, G. Passante, and B. Wilcox, “Disparities in access to U.S. quantum information education,”Physical Review Physics Education Research, vol. 20, p. 010131, May 2024

    work page 2024

  3. [3]

    IBM Quantum Computing,

    IBM Quantum, “IBM Quantum Computing,” https://www.ibm. com/quantum, 2024

    work page 2024

  4. [4]

    National quantum initiative act,

    115th U.S. Congress, “National quantum initiative act,” United States Congress, Washington, DC, 2018

    work page 2018

  5. [5]

    Toward personalizing quan- tum computing education: An evolutionary LLM-powered ap- proach,

    I. Elhaimeur and N. Chrisochoides, “Toward personalizing quan- tum computing education: An evolutionary LLM-powered ap- proach,” inProceedings of the IEEE International Conference on Quantum Computing and Engineering (QCE), 2025

    work page 2025

  6. [6]

    Understanding quantum information and computa- tion,

    J. Watrous, “Understanding quantum information and computa- tion,” 2025

    work page 2025

  7. [7]

    Latency and cost of multi- agent intelligent tutoring at scale,

    I. Elhaimeur and N. Chrisochoides, “Latency and cost of multi- agent intelligent tutoring at scale,” 2026

    work page 2026

  8. [8]

    Itas: A multi-agent architecture for llm-based intelligent tutoring,

    ——, “Itas: A multi-agent architecture for llm-based intelligent tutoring,” 2026

    work page 2026

  9. [9]

    AutoTutor: A tutor with dialogue in natural language,

    A. C. Graesser, S. Lu, G. T. Jackson, H. H. Mitchell, M. Ven- tura, A. Olney, and M. M. Louwerse, “AutoTutor: A tutor with dialogue in natural language,”Behavior Research Methods, Instruments, & Computers, vol. 36, no. 2, pp. 180–193, 2004

    work page 2004

  10. [10]

    LLM agents for education: Advances and applications,

    Z. Chu, S. Wang, J. Xie, T. Zhu, Y . Yan, J. Ye, A. Zhong, X. Hu, J. Liang, P. S. Yu, and Q. Wen, “LLM agents for education: Advances and applications,” 2025

    work page 2025

  11. [11]

    Towards a science of scaling agent systems,

    Y . Kim, K. Gu, C. Park, C. Park, S. Schmidgall, A. A. Heydari, Y . Yan, Z. Zhang, Y . Zhuang, M. Malhotra, P. P. Liang, H. W. Park, Y . Yang, X. Xu, Y . Du, S. Patel, T. Althoff, D. McDuff, and X. Liu, “Towards a science of scaling agent systems,” 2025

    work page 2025

  12. [12]

    AutoGen: Enabling next-gen LLM applications via multi-agent conversation,

    Q. Wu, G. Bansal, J. Zhang, Y . Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liu, A. H. Awadallah, R. W. White, D. Burger, and C. Wang, “AutoGen: Enabling next-gen LLM applications via multi-agent conversation,” inProceedings of the Third Conference on Language Modeling (COLM 2024), 2024

    work page 2024

  13. [13]

    MetaGPT: Meta program- ming for a multi-agent collaborative framework,

    S. Hong, M. Zhuge, J. Chen, X. Zheng, Y . Cheng, J. Wang, C. Zhang, Z. Wang, S. K. S. Yau, Z. Lin, L. Zhou, C. Ran, L. Xiao, C. Wu, and J. Schmidhuber, “MetaGPT: Meta program- ming for a multi-agent collaborative framework,” inProceedings of the Twelfth International Conference on Learning Represen- tations (ICLR 2024), 2024

    work page 2024

  14. [14]

    Understanding quantum information and computation: IBM quantum learning lecture series,

    J. Watrous, “Understanding quantum information and computation: IBM quantum learning lecture series,” https://learning.quantum.ibm.com/course/ understanding-quantum-information-and-computation, 2024

    work page 2024

  15. [15]

    Investigating and improving student understanding of the basics of quantum computing,

    P. Hu, Y . Li, and C. Singh, “Investigating and improving student understanding of the basics of quantum computing,”Physical Review Physics Education Research, vol. 20, p. 020108, 2024

    work page 2024

  16. [16]

    Addressing misconceptions in university physics: A review and experiences from quantum physics educators,

    S. Majidy, “Addressing misconceptions in university physics: A review and experiences from quantum physics educators,” 2025

    work page 2025

  17. [17]

    Experi- ence in teaching quantum computing with hands-on programming labs,

    F. Galetto, H. H. L ´opez, M. Rahmati, J. Sang, and C. Yu, “Experi- ence in teaching quantum computing with hands-on programming labs,”The Journal of Supercomputing, vol. 80, pp. 14 029–14 056, 2024

    work page 2024

  18. [18]

    How video production affects student engagement: An empirical study of MOOC videos,

    P. J. Guo, J. Kim, and R. Rubin, “How video production affects student engagement: An empirical study of MOOC videos,” in Proceedings of the First ACM Conference on Learning @ Scale (L@S ’14), 2014, pp. 41–50

    work page 2014

  19. [19]

    Programming pluralism: Using learning analytics to detect patterns in the learning of computer programming,

    P. Blikstein, M. Worsley, C. Piech, M. Sahami, S. Cooper, and D. Koller, “Programming pluralism: Using learning analytics to detect patterns in the learning of computer programming,” Journal of the Learning Sciences, vol. 23, no. 4, pp. 561–599, 2014

    work page 2014

  20. [20]

    Multimodal learning analytics and education data mining: Using computational technologies to measure complex learning tasks,

    P. Blikstein and M. Worsley, “Multimodal learning analytics and education data mining: Using computational technologies to measure complex learning tasks,”Journal of Learning Analytics, vol. 3, no. 2, pp. 220–238, 2016

    work page 2016

  21. [21]

    Deconstructing disengagement: Analyzing learner subpopulations in massive open online courses,

    R. F. Kizilcec, C. Piech, and E. Schneider, “Deconstructing disengagement: Analyzing learner subpopulations in massive open online courses,” inProceedings of the Third International Conference on Learning Analytics and Knowledge (LAK ’13), 2013, pp. 170–179

    work page 2013

  22. [22]

    Education for expanding the quantum workforce: Students’ perceptions of the quantum industry in an upper-division physics capstone course,

    K. A. Oliver, V . Borish, B. R. Wilcox, and H. J. Lewandowski, “Education for expanding the quantum workforce: Students’ perceptions of the quantum industry in an upper-division physics capstone course,”Physical Review Physics Education Research, vol. 21, p. 010129, 2025

    work page 2025

  23. [23]

    Enabling large-scale quantum computing via dis- tributed and hybrid architectures,

    W. Tang, “Enabling large-scale quantum computing via dis- tributed and hybrid architectures,” Ph.D. dissertation, Princeton University, 2025

    work page 2025

  24. [24]

    Azfar, R

    A. Maciejunes, J. Stenger, D. Gunlycke, and N. Chrisochoides, “Solving large-scale vehicle routing problems with hybrid quantum-classical decomposition,”arXiv preprint arXiv:2507.05373, 2025

    work page arXiv 2025

  25. [25]

    Towards a utility-scale quantum edge detection for real-world medical image data,

    E. Billias and N. Chrisochoides, “Towards a utility-scale quantum edge detection for real-world medical image data,”arXiv preprint arXiv:2507.10939, 2025

    work page arXiv 2025

  26. [26]

    Teaching quantum computing with an interactive textbook,

    J. R. Wootton, F. Harkins, N. T. Bronn, A. Carrera Vazquez, A. Phan, and A. T. Asfaw, “Teaching quantum computing with an interactive textbook,” in2021 IEEE International Conference on Quantum Computing and Engineering (QCE), 2021, pp. 385– 391

    work page 2021

  27. [27]

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,

    L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qin, and T. Liu, “A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions,”ACM Transactions on Information Systems, 2024

    work page 2024

  28. [28]

    MonarchSphere: AI incubator powered by Google Cloud,

    Old Dominion University, “MonarchSphere: AI incubator powered by Google Cloud,” https://www.odu.edu/ forward-focused-transformation/monarchsphere, 2025

    work page 2025

  29. [29]

    Hohpe and B

    G. Hohpe and B. Woolf,Enterprise Integration Patterns: Design- ing, Building, and Deploying Messaging Solutions. Addison- Wesley, 2003

    work page 2003

  30. [30]

    Family educational rights and privacy act (FERPA),

    U.S. Congress, “Family educational rights and privacy act (FERPA),” 20 U.S.C. § 1232g; 34 CFR Part 99, 1974

    work page 1974

  31. [31]

    Quantum computing and large language models: An overview,

    R. Kharsa, A. Bouridane, and A. Abadla, “Quantum computing and large language models: An overview,” in2024 International Conference on Electrical, Computer and Energy Technologies (ICECET), 2024

    work page 2024

  32. [32]

    Introductory quantum information science coursework at US in- stitutions: content coverage,

    J. C. Meyer, G. Passante, S. J. Pollock, and B. R. Wilcox, “Introductory quantum information science coursework at US in- stitutions: content coverage,”EPJ Quantum Technology, vol. 11, no. 1, p. 16, 2024

    work page 2024

  33. [33]

    Today’s interdisciplinary quantum information classroom: Themes from a survey of quantum information science instruc- tors,

    ——, “Today’s interdisciplinary quantum information classroom: Themes from a survey of quantum information science instruc- tors,”Physical Review Physics Education Research, vol. 18, p. 010150, 2022

    work page 2022

  34. [34]

    M. Q. Patton,Qualitative Research and Evaluation Methods, 3rd ed. Thousand Oaks, CA: Sage Publications, 2002

    work page 2002

  35. [35]

    S. B. Merriam,Qualitative Research: A Guide to Design and Implementation. San Francisco, CA: Jossey-Bass, 2009

    work page 2009

  36. [36]

    Adaptable curricular exercises for QIS (ACE-QIS),

    G. Passante, B. R. Wilcox, S. J. Pollock, and G. Corsiglia, “Adaptable curricular exercises for QIS (ACE-QIS),” PhysPort, https://www.physport.org/curricula/ACEQIS/, 2023

    work page 2023

  37. [37]

    Development and uses of upper-division conceptual assessments,

    B. R. Wilcox, M. D. Caballero, C. Baily, H. Sadaghiani, S. V . Chasteen, Q. X. Ryan, and S. J. Pollock, “Development and uses of upper-division conceptual assessments,”Physical Review Special Topics – Physics Education Research, vol. 11, p. 020115, 2015

    work page 2015