pith. sign in

arxiv: 2605.05144 · v1 · submitted 2026-05-06 · 💻 cs.LG · cs.CY

Human-AI Co-Mentorship in Project-Based Learning: A Case Study in Financial Forecasting

Freyaa Chawla , Ahan Chawla , Rishi Singh , Joe Germino , Grigorii Khvatskii This is my paper

Pith reviewed 2026-05-08 16:40 UTC · model grok-4.3

classification 💻 cs.LG cs.CY
keywords AI mentorshipproject-based learningfinancial forecastinghigh school educationmachine learning educationETF predictionworkflow design
0
0 comments X

The pith

High school students with limited AI and finance background built functional ETF forecasting models by designing workflows and using AI tools for execution.

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

This paper presents a case study of a summer project in which high-school and early-undergraduate students, mentored by graduate researchers, tackled financial forecasting for ETF prices. Students began by identifying the sequence of steps required for the problem and then relied on AI tools to generate and refine the code for each step. Daily stand-up meetings served to clarify concepts, debug issues, and keep the group aligned. The students, who entered with minimal technical preparation, produced models with real-world applicability while each pursued personal interests in computer science or finance. The work shows how AI assistance can shift the focus of project-based learning toward higher-order formulation rather than sequential skill acquisition.

Core claim

Students identified the overall workflow for market analysis and ETF price prediction, then iterated with AI tools to implement the technical components while using daily stand-ups with graduate mentors to handle conceptual questions and debugging. This process enabled each participant to deepen their engagement in their chosen area and produced meaningful forecasting models despite the team's limited prior qualifications in AI or finance.

What carries the argument

AI-assisted workflow design in which students first define the sequence of problem-solving steps and then delegate execution to AI tools, with human mentorship limited to stand-ups for clarification and debugging.

If this is right

  • Students shift from learning methods in isolation to applying them immediately within a defined workflow.
  • Individual interests in computer science or finance can be pursued while still advancing a shared technical deliverable.
  • Meaningful models applicable to real financial tasks can be completed in one summer despite minimal starting qualifications.
  • Daily stand-ups become the main site for conceptual learning and problem definition rather than for teaching foundational skills.

Where Pith is reading between the lines

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

  • The same pattern of workflow-first design followed by AI execution could be tested in other technical domains such as biology or environmental science to see whether early students can reach similar outcomes.
  • Programs might explore whether reducing the prerequisite teaching load allows more time for students to tackle open-ended or interdisciplinary questions.
  • Over-reliance on AI execution could be checked by measuring whether students retain the ability to explain or modify the generated code without further tool assistance.

Load-bearing premise

The assumption that the students' success in building working models came primarily from the AI tools and workflow structure rather than from the graduate researchers' guidance or the students' own enthusiasm and capabilities.

What would settle it

A parallel project with a comparable group of students who receive the same graduate mentorship and project goal but lack access to AI coding tools, and who fail to produce functional forecasting models within the same timeframe.

Figures

Figures reproduced from arXiv: 2605.05144 by Ahan Chawla, Freyaa Chawla, Grigorii Khvatskii, Joe Germino, Rishi Singh.

Figure 1
Figure 1. Figure 1: Price delta prediction example (BBJP) Additionally, it should be noted that in all cases, the models that included the sentiment as one of their inputs underperformed when compared to their counterparts that had no access to the sentiment information. We hypothesize two explanations: first, sentiment signals may already be incorporated into price movements by the time they are observable in news; second, t… view at source ↗
read the original abstract

This paper reflects on a AI research project carried out by a team of high-school and early-undergraduate students under the mentorship of graduate researchers and ably assisted by AI tools. We share our experience in not only on the learning experience for the high school students, but also on how AI tools accelerated the process that enabled the high school students to focus on higher order problem formulation and solution. Although the participants entered the project with limited background in both AI and finance, they showed strong enthusiasm for technical market analysis and ETF price prediction. Traditional learning settings would first teach the necessary methods in a classroom setting and only later let students apply them. In contrast, our project emphasized workflow design: students identified the sequence of steps needed to address the problem and then used AI-driven tools to execute each step. We note that the high school students developed the necessary code through iterating with the AI tools, and we used our daily stand-ups to debug and answer conceptual questions. Each of the student was able to dig deeper into their area of interest whether computer science or finance, while collaboratively making a significant advance over the summer of 2025. This project was an important pedagogical exercise on how AI tools can be used for mentoring high school students, allowing them to focus on their specific interests and using the daily stand-ups to focus on problem definition and conceptual understanding. Despite their limited technical qualifications, the students were able to leverage AI tools to build meaningful models with real-world application.

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

3 major / 2 minor

Summary. The paper reports on a project-based learning initiative in financial forecasting involving high-school and early-undergraduate students mentored by graduate researchers with AI tool assistance. Students used a workflow design method to identify steps for ETF price prediction and leveraged AI for code and implementation, with daily stand-ups for support. The authors conclude that this human-AI co-mentorship enabled students with limited technical backgrounds to develop meaningful models with real-world applications and engage in higher-order problem solving.

Significance. Should the described outcomes be verified through additional evidence, this case study could demonstrate the potential of AI tools to democratize access to complex technical projects in education. It offers a model for shifting from traditional instruction to integrated workflow and AI-supported application, which may influence future designs for project-based learning in AI and finance domains.

major comments (3)
  1. Abstract: The assertion that 'the students were able to leverage AI tools to build meaningful models with real-world application' lacks any concrete details or evidence regarding the models, such as the specific data sources for ETF prices, the machine learning techniques employed, evaluation metrics, or how 'real-world application' was determined. This absence directly undermines the paper's primary claim about the success of the AI-assisted approach.
  2. Abstract: The manuscript does not provide any metrics or measures to evaluate the effectiveness of the AI co-mentorship, such as model performance results, student skill development assessments, or comparisons with traditional mentorship methods. This makes the pedagogical insights anecdotal and difficult to generalize.
  3. Abstract: No consideration is given to alternative explanations for the students' achievements, including the direct input from graduate mentors or pre-existing student motivations, which could be addressed in a dedicated limitations or discussion section to strengthen the attribution to the AI workflow design.
minor comments (2)
  1. Abstract: The phrasing 'ably assisted by AI tools' and 'we used our daily stand-ups' mixes first-person perspectives without clear delineation of author roles.
  2. The paper would benefit from including specific examples of the steps in the workflow design or sample interactions with AI tools to illustrate the process described.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our case study. We agree that the abstract requires clarification and that a limitations section would strengthen the manuscript. Below we respond to each major comment. We will revise the abstract to better align with the reflective scope of the work and add a dedicated limitations/discussion section.

read point-by-point responses

  1. Referee: The assertion that 'the students were able to leverage AI tools to build meaningful models with real-world application' lacks any concrete details or evidence regarding the models, such as the specific data sources for ETF prices, the machine learning techniques employed, evaluation metrics, or how 'real-world application' was determined. This absence directly undermines the paper's primary claim about the success of the AI-assisted approach.

    Authors: We acknowledge the abstract is overly broad. The manuscript is a reflective case study centered on the human-AI co-mentorship and workflow design process, not a technical evaluation of forecasting models. No specific data sources, ML techniques, or performance metrics are reported because the project did not track or emphasize these; 'meaningful' and 'real-world application' refer to students successfully implementing a basic workflow on actual ETF price data for learning purposes. We will revise the abstract to qualify this claim, provide a high-level description of the workflow steps, and remove language that implies technical success or generalizability. revision: yes

  2. Referee: The manuscript does not provide any metrics or measures to evaluate the effectiveness of the AI co-mentorship, such as model performance results, student skill development assessments, or comparisons with traditional mentorship methods. This makes the pedagogical insights anecdotal and difficult to generalize.

    Authors: As a single-project qualitative case study, the paper relies on observational evidence from daily stand-ups and student engagement rather than formal metrics, assessments, or control-group comparisons. We agree this renders the insights anecdotal. We will add a limitations section explicitly stating the absence of quantitative evaluation and the challenges of generalizing from this format, while retaining the value of the detailed process description. revision: yes

  3. Referee: No consideration is given to alternative explanations for the students' achievements, including the direct input from graduate mentors or pre-existing student motivations, which could be addressed in a dedicated limitations or discussion section to strengthen the attribution to the AI workflow design.

    Authors: We agree that the roles of graduate mentor guidance during stand-ups and students' self-selected interest and motivation are plausible contributing factors. We will add a limitations paragraph in the discussion section that addresses these alternatives and clarifies how the AI workflow and co-mentorship interacted with them, without overstating the unique contribution of the AI component. revision: yes

Circularity Check

0 steps flagged

No circularity: purely descriptive case study with no derivations or predictions

full rationale

The manuscript is a reflective narrative on a student AI-mentorship project in financial forecasting. It contains no equations, no fitted parameters, no claimed predictions, and no derivation chain that could reduce to its own inputs. The central claim (students built meaningful models) is presented as an observational outcome of the described workflow, without any self-referential logic, self-citation load-bearing premises, or renaming of known results. This is a standard non-finding for qualitative experience reports; the absence of technical derivations makes circularity impossible by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is a qualitative case study with no quantitative models, mathematical derivations, or technical postulates; it introduces no free parameters, axioms, or invented entities.

pith-pipeline@v0.9.0 · 5585 in / 1136 out tokens · 96823 ms · 2026-05-08T16:40:06.883440+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

19 extracted references · 7 canonical work pages

  1. [1]

    The wicked problem of ai and assessment,

    T. Corbin, M. Bearman, D. Boud, and P. Dawson, “The wicked problem of ai and assessment,”Assessment & evaluation in higher education, pp. 1–17, 2025

    2025

  2. [2]

    Towards robust evaluation of stem education: Leveraging mllms in project-based learning,

    X. Wu, Y. Jia, Q. Zhang, Y. Qin, L. Xiao, and S. Zhao, “Towards robust evaluation of stem education: Leveraging mllms in project-based learning,”arXiv preprint arXiv:2505.17050, 2025

    work page arXiv 2025

  3. [3]

    Evaluation of llm-based feedback generation for distance project-based learning,

    K. Sasaki and T. Inoue, “Evaluation of llm-based feedback generation for distance project-based learning,” inInternational Conference on Collaboration Technologies and Social Computing, Springer, 2025, pp. 144–160

    2025

  4. [4]

    Automatic feedback generation on k-12 students’ data science education by prompting cloud-based large language models,

    S. C. E. Fung, M. F. Wong, and C. W. Tan, “Automatic feedback generation on k-12 students’ data science education by prompting cloud-based large language models,” in Proceedings of the Eleventh ACM Conference on Learning@ Scale, 2024, pp. 255–258

    2024

  5. [5]

    Autopbl: An llm-powered platform to guide and support individual learners through self project-based learning,

    Y. Zhu, Z. Ye, Y. Yuan, W. Tang, C. Yu, and Y. Shi, “Autopbl: An llm-powered platform to guide and support individual learners through self project-based learning,” inProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025, pp. 1–26

    2025

  6. [6]

    Design and evaluation of an llm-based mentor for software architecture in higher education project management classes,

    S. G¨ urtl, D. Scharf, C. Thrainer, C. G¨ utl, and A. Steinmaurer, “Design and evaluation of an llm-based mentor for software architecture in higher education project management classes,” inInternational Conference on Interactive Collaborative Learning, Springer, 2024, pp. 375–386

    2024

  7. [7]

    Enhancing statistics education through project-based learning (pbl) and the emergence of chatgpt,

    L. Al Labadi and A. Ly, “Enhancing statistics education through project-based learning (pbl) and the emergence of chatgpt,”Teaching Statistics, vol. 47, no. 3, pp. 200–218, 2025. doi:https://doi.org/10.1111/test.12405eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/test.12405. [Online]. Available:https://onlinelibrary.wiley.com/doi/abs/10.1111/test.12405

    work page doi:10.1111/test.12405eprint: 2025

  8. [8]

    Application of generative artificial intelligence to assessment and curriculum design for project-based learning,

    T. Wu and M. Chang, “Application of generative artificial intelligence to assessment and curriculum design for project-based learning,” in2023 International Conference on Engineering and Emerging Technologies (ICEET), 2023, pp. 1–6.doi: 10.1109/ICEET60227.2023.10525933

    work page doi:10.1109/iceet60227.2023.10525933 2023

  9. [9]

    Cs1-llm: Integrating llms into cs1 instruction,

    A. Vadaparty et al., “Cs1-llm: Integrating llms into cs1 instruction,” inProceedings of the 2024 on Innovation and Technology in Computer Science Education V. 1, ser. ITiCSE 2024, Milan, Italy: Association for Computing Machinery, 2024, pp. 297–303,isbn: 9798400706004.doi:10.1145/3649217.3653584[Online]. Available: https://doi.org/10.1145/3649217.3653584

    work page doi:10.1145/3649217.3653584 2024

  10. [10]

    Integrating large language models into project-based learning based on self-determination theory,

    L. Zhang and W. Zhang, “Integrating large language models into project-based learning based on self-determination theory,”Interactive Learning Environments, vol. 33, no. 5, pp. 3580–3592, 2025

    2025

  11. [11]

    S. Zha, Y. Qiao, Q. Hu, Z. Li, J. Gong, and Y. Xu,Designing child-centric ai learning environments: Insights from llm-enhanced creative project-based learning, 2024. arXiv: 2403.16159 [cs.HC]. [Online]. Available:https://arxiv.org/abs/2403.16159

    work page arXiv 2024

  12. [12]

    A two-level cascade evolutionary computation based covered call trading model,

    M. Ucar, I. Bayram, and A. M. Ozbayoglu, “A two-level cascade evolutionary computation based covered call trading model,”Procedia Computer Science, vol. 20, pp. 472–477, 2013

    2013

  13. [13]

    Machine learning method for return direction forecast of exchange traded funds (etfs) using classification and regression models,

    R. P. B. Piovezan, P. P. de Andrade Junior, and S. L. ´Avila, “Machine learning method for return direction forecast of exchange traded funds (etfs) using classification and regression models,”Computational Economics, vol. 63, no. 5, pp. 1827–1852, 2024

    2024

  14. [14]

    Multimodel forecasting of technology-focused etfs using arima and lstm,

    S. Sharma, U. Anand, V. Aggarwal, P. Jain, and A. Anand, “Multimodel forecasting of technology-focused etfs using arima and lstm,” in2025 3rd International Conference on Self Sustainable Artificial Intelligence Systems (ICSSAS), IEEE, 2025, pp. 559–565

    2025

  15. [15]

    Xgboost algorithm-based model for predicting mutual fund returns,

    X. Li, L. Yang, C. Zha, and Y. Xu, “Xgboost algorithm-based model for predicting mutual fund returns,”Computational Economics, pp. 1–26, 2026

    2026

  16. [16]

    Predicting daily stock price directions with deep learning models,

    T. Kundu and E. Pinsky, “Predicting daily stock price directions with deep learning models,”Machine Learning with Applications, vol. 22, p. 100 744, 2025

    2025

  17. [17]

    A comparative study of lstm, lightgbm, and autoregressive model in narrow-based etf market prediction with multi-ticker models,

    M. Hong, Z. Chen, W. Mahmoud Soliman, and K. Zhang, “A comparative study of lstm, lightgbm, and autoregressive model in narrow-based etf market prediction with multi-ticker models,” inProceedings of the 6th International Conference on Machine Learning and Machine Intelligence, ser. MLMI ’23, Chongqing, China: Association for Computing Machinery, 2024, pp....

    work page doi:10.1145/3635638.3635640 2024

  18. [18]

    Aroussi,Ranaroussi/yfinance, original-date: 2017-05-21T10:16:15Z, Jan

    R. Aroussi,Ranaroussi/yfinance, original-date: 2017-05-21T10:16:15Z, Jan. 2026. Accessed: Jan. 21, 2026. [Online]. Available: https://github.com/ranaroussi/yfinance

    2017

  19. [19]

    Mean Square Error,

    M. D. Schluchter, “Mean Square Error,” en, inWiley StatsRef: Statistics Reference Online, R. S. Kenett, N. T. Longford, W. W. Piegorsch, and F. Ruggeri, Eds., 1st ed., Wiley, Sep. 2014,isbn: 9781118445112.doi:10.1002/9781118445112.stat05906Accessed: Jan. 21, 2026. [Online]. Available: https://onlinelibrary.wiley.com/doi/10.1002/9781118445112.stat05906

    work page doi:10.1002/9781118445112.stat05906accessed: 2014