pith. sign in

arxiv: 2605.05842 · v1 · submitted 2026-05-07 · 💻 cs.AI

Taklif.AI: LLM-Powered Platform for Interest-Based Personalized College Assignments

Zaki Kurdya , Mohammed Zuqlam , Salem Amassi , Shady Telbany , Motaz Saad This is my paper

Pith reviewed 2026-05-08 11:17 UTC · model grok-4.3

classification 💻 cs.AI
keywords personalized assignmentslarge language modelsprompt engineeringeducational technologystudent engagementAI guardrailscultural contextcollege education
0
0 comments X

The pith

Taklif.AI generates college assignments personalized to students' extracurricular interests and cultural contexts using large language models with prompt guardrails.

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

Educators often rely on uniform assignments that overlook students' varied personal interests, which can lower motivation and encourage copying. Taklif.AI counters this by routing student-provided interest and background details through large language models via a fixed prompt pipeline. Input and output guardrails are added to keep the resulting tasks educationally appropriate and free of obvious errors. Early feedback from 68 users showed that most viewed the interest-matching feature as useful. The platform therefore offers a concrete way to move assignment design from generic templates toward individual relevance.

Core claim

Taklif.AI is a platform that leverages large language models to automatically generate personalized assignments tailored to individual student interests and cultural contexts through a structured prompt engineering pipeline with input and output guardrails, deployed on a serverless AWS architecture using Next.js, Llama 3.3 70B via LiteLLM, and LangChain for orchestration, with preliminary user acceptance testing indicating that 84 percent of 68 participants rated the personalization feature as beneficial.

What carries the argument

The structured prompt engineering pipeline with input and output guardrails that folds extracurricular interests and cultural contexts into assignment generation while enforcing quality checks.

If this is right

  • Assignments matched to students' own interests can replace one-size-fits-all tasks and thereby raise participation.
  • Guardrails placed around the language-model output can limit off-topic or inaccurate content in the generated work.
  • A serverless cloud setup combined with multi-provider model access supports practical scaling for classroom use.
  • Positive early user ratings from both students and educators point toward feasible adoption in higher education settings.
  • Directions for later controlled studies are needed to confirm effects on actual learning results.

Where Pith is reading between the lines

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

  • The same interest-injection method could be tested on K-12 assignments or professional training materials where motivation is also a problem.
  • Unique per-student tasks may lower incentives for plagiarism even before any plagiarism-detection tools are applied.
  • Cultural-context inclusion carries a risk of embedding subtle stereotypes that would require separate auditing beyond the current guardrails.
  • The guardrail pattern itself might transfer to other LLM uses such as generating individualized feedback or study plans.

Load-bearing premise

Prompt-based integration of personal interests via large language models will produce educationally sound assignments that improve engagement without introducing bias, hallucinations, or reduced learning value.

What would settle it

A side-by-side trial that tracks actual student engagement and learning gains on assignments generated by the platform versus standard assignments and finds no measurable benefit or frequent quality failures.

Figures

Figures reproduced from arXiv: 2605.05842 by Mohammed Zuqlam, Motaz Saad, Salem Amassi, Shady Telbany, Zaki Kurdya.

Figure 1
Figure 1. Figure 1: Overall four-layer architecture of Taklif.AI. User requests flow from the Web Layer through the API Layer, view at source ↗
Figure 2
Figure 2. Figure 2: Third-Party AI Layer detail. Requests are routed through LangChain for prompt structuring, then to LiteLLM view at source ↗
Figure 3
Figure 3. Figure 3: Input parsing and task routing. The handler function extracts the task type and routes to the appropriate parser view at source ↗
Figure 4
Figure 4. Figure 4: Interest guardrails flow. Student interests are validated before LLM processing. Invalid interests trigger a 400 view at source ↗
Figure 5
Figure 5. Figure 5: Output guardrails and response handling. AI-generated content is validated for safety and relevance before view at source ↗
read the original abstract

Educators face significant challenges in creating engaging, personalized assignments that accommodate students' diverse interests and cognitive abilities. Traditional one-size-fits-all assignments frequently lead to decreased student engagement and increased reliance on unethical practices such as plagiarism. To address these challenges, we present Taklif.AI, a platform that leverages Large Language Models (LLMs) to automatically generate personalized assignments tailored to individual student interests. Unlike existing AI-powered educational platforms that personalize based on academic performance metrics alone, Taklif.AI incorporates students' extracurricular interests and cultural contexts into the assignment generation process through a structured prompt engineering pipeline with input and output guardrails. The platform employs a serverless architecture on AWS with Next.js, using Llama 3.3 70B as the primary LLM via LiteLLM for multi-provider load balancing and LangChain for prompt orchestration. We describe the system architecture, the prompt design methodology, and the guardrails framework that ensures output quality. Preliminary user acceptance testing with 68 participants (65 students and 3 educators) indicates positive reception, with 84% of participants rating the personalization feature as beneficial. We discuss the platform's current capabilities and limitations, and outline directions for rigorous empirical evaluation of learning outcomes.

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 / 2 minor

Summary. The manuscript presents Taklif.AI, an LLM-powered platform for generating interest-based personalized college assignments. It uses a structured prompt engineering pipeline with input and output guardrails to incorporate students' extracurricular interests and cultural contexts, implemented via a serverless AWS architecture with Next.js, Llama 3.3 70B via LiteLLM, and LangChain for orchestration. The paper describes the system architecture, prompt design methodology, guardrails framework, and reports preliminary user acceptance testing with 68 participants (65 students, 3 educators) where 84% rated the personalization feature as beneficial.

Significance. If the described prompt pipeline and guardrails reliably produce educationally sound assignments, the platform could meaningfully advance interest-based personalization in higher education beyond performance metrics alone, potentially improving engagement and reducing plagiarism. The work provides a concrete system implementation with open technical details on architecture and orchestration, which is a strength for reproducibility in applied AI education tools. However, the current evidence base is limited to subjective acceptance ratings without demonstrated effects on learning outcomes.

major comments (2)
  1. [User Acceptance Testing / Abstract] The preliminary user acceptance testing (described in the abstract and evaluation section) reports 84% positive feedback on personalization from 68 participants but provides no methodology details, survey instrument, sampling procedure, controls, statistical analysis, or breakdown by student vs. educator responses. This leaves the central claim of positive reception without sufficient support for assessing reliability or generalizability.
  2. [Guardrails Framework] The guardrails framework is presented as ensuring output quality and preventing issues such as hallucinations or bias, yet the manuscript contains no empirical evaluation of guardrail effectiveness (e.g., expert review of sample assignments, hallucination rates, bias audits, or comparison against baseline LLM outputs without guardrails). This is load-bearing for the claim that the structured pipeline produces educationally sound assignments.
minor comments (2)
  1. [Abstract] The abstract states that future work will include 'rigorous empirical evaluation of learning outcomes' but does not outline specific metrics, study designs, or control conditions planned for that evaluation.
  2. [Prompt Design Methodology] The description of the prompt engineering pipeline would benefit from an example of a full input prompt template and corresponding output to illustrate how extracurricular interests are concretely integrated.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments on our manuscript. We address each major comment below, indicating revisions where we agree changes are warranted to improve clarity and transparency.

read point-by-point responses

  1. Referee: [User Acceptance Testing / Abstract] The preliminary user acceptance testing (described in the abstract and evaluation section) reports 84% positive feedback on personalization from 68 participants but provides no methodology details, survey instrument, sampling procedure, controls, statistical analysis, or breakdown by student vs. educator responses. This leaves the central claim of positive reception without sufficient support for assessing reliability or generalizability.

    Authors: We agree that the description of the user acceptance testing requires more detail to allow proper assessment. In the revised manuscript, we will expand the evaluation section to specify the survey instrument (a 5-point Likert-scale questionnaire on the perceived benefits of interest-based personalization), sampling procedure (convenience sampling through university student and educator networks), administration (participants completed a platform demo prior to responding), and response breakdown (85% positive among the 65 students and 67% among the 3 educators). We will also note the absence of statistical controls or formal analysis, given the preliminary exploratory nature of the testing, and list this explicitly as a limitation. These additions will not change the preliminary status of the results but will address concerns about transparency. revision: yes

  2. Referee: [Guardrails Framework] The guardrails framework is presented as ensuring output quality and preventing issues such as hallucinations or bias, yet the manuscript contains no empirical evaluation of guardrail effectiveness (e.g., expert review of sample assignments, hallucination rates, bias audits, or comparison against baseline LLM outputs without guardrails). This is load-bearing for the claim that the structured pipeline produces educationally sound assignments.

    Authors: We acknowledge that the manuscript does not include quantitative empirical evaluation of the guardrails, such as hallucination rates or expert audits. The primary focus of this work is the system design, architecture, and prompt pipeline rather than a controlled evaluation study. In revision, we will augment the guardrails section with concrete implementation examples (e.g., specific input validation for cultural context and output filters for factual consistency and bias) and explicitly state the lack of such empirical validation as a limitation, while outlining plans for future work including baseline comparisons. We maintain that the preliminary user feedback provides initial supporting evidence for the overall approach, but we accept that stronger validation would be valuable in subsequent studies. revision: partial

Circularity Check

0 steps flagged

No circularity: descriptive system implementation with no derivations or load-bearing self-citations

full rationale

The paper is a system description of Taklif.AI, detailing LLM-based assignment generation via prompt engineering, AWS serverless architecture, LangChain orchestration, and guardrails. It reports preliminary user acceptance from 68 participants (84% positive on personalization) but makes no mathematical claims, predictions, fitted parameters, or derivations. No equations, uniqueness theorems, or self-citation chains appear. The central content is architectural and methodological description plus initial survey results, which are self-contained and do not reduce to inputs by construction. This matches the default expectation of no significant circularity for implementation papers.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The platform rests on standard assumptions about LLM text generation capabilities and prompt effectiveness; no free parameters, new axioms, or invented entities are introduced in the abstract.

pith-pipeline@v0.9.0 · 5523 in / 1063 out tokens · 29850 ms · 2026-05-08T11:17:23.523291+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

41 extracted references · 41 canonical work pages · 2 internal anchors

  1. [1]

    Walkington

    Candace A. Walkington. Using adaptive learning technologies to personalize instruction to student interests: The impact of relevant contexts on performance and learning outcomes.Journal of Educational Psychology, 105(4):932–945, 2013

    work page 2013

  2. [2]

    Eliminating plagiarism in programming courses through assessment design.International Journal of Information and Education Technology, 6(11):873–879, Nov 2016

    Minh Ngoc Ngo. Eliminating plagiarism in programming courses through assessment design.International Journal of Information and Education Technology, 6(11):873–879, Nov 2016

    work page 2016

  3. [3]

    CS for all: Catering to diversity of master’s students through assignment choices

    Sohail Alhazmi, Margaret Hamilton, and Charles Thevathayan. CS for all: Catering to diversity of master’s students through assignment choices. InProceedings of the 49th ACM Technical Symposium on Computer Science Education (SIGCSE ’18), pages 38–43. Association for Computing Machinery, 2018

    work page 2018

  4. [4]

    The role of students’ interests during computer-assisted learning: A field experiment.Computers in Human Behavior, 130:107168, 2022

    Kaat Iterbeke, Wouter Schelfhout, and Kristof De Witte. The role of students’ interests during computer-assisted learning: A field experiment.Computers in Human Behavior, 130:107168, 2022

    work page 2022

  5. [5]

    New era of artificial intelligence in education: Towards a sustainable multifaceted revolution.Sustainability, 15(16):12451, 2023

    Firuz Kamalov, David Santandreu Calonge, and Ikhlaas Gurrib. New era of artificial intelligence in education: Towards a sustainable multifaceted revolution.Sustainability, 15(16):12451, 2023

    work page 2023

  6. [6]

    Education technology: An evidence- based review

    Maya Escueta, Vincent Quan, Andre Joshua Nickow, and Philip Oreopoulos. Education technology: An evidence- based review. Working Paper 23744, National Bureau of Economic Research, 2017

    work page 2017

  7. [7]

    AI literacy in K-12: A systematic literature review.International Journal of STEM Education, 10(1):29, 2023

    Lorena Casal-Otero, Alejandro Catala, Carmen Fernández-Morante, Maria Taboada, Beatriz Cebreiro, and Senén Barro. AI literacy in K-12: A systematic literature review.International Journal of STEM Education, 10(1):29, 2023

    work page 2023

  8. [8]

    Alamri, Victoria Lynn Lowell, William R

    Hamdan A. Alamri, Victoria Lynn Lowell, William R. Watson, and Sunnie Lee Watson. Using personalized learning as an instructional approach to motivate learners in online higher education: Learner self-determination and intrinsic motivation.Journal of Research on Technology in Education, 52:322 – 352, 2020

    work page 2020

  9. [9]

    A comprehensive analysis of personalized learning components.Journal of Computers in Education, 8(4):485–503, 2021

    Atikah Shemshack, Kinshuk, and Jonathan Michael Spector. A comprehensive analysis of personalized learning components.Journal of Computers in Education, 8(4):485–503, 2021

    work page 2021

  10. [10]

    Generation of student interest in an inquiry-based mobile learning environment.Frontline Learning Research, 5(4):42–60, 2017

    Erkka Laine, Marjaana Veermans, Aleksi Lahti, and Koen Veermans. Generation of student interest in an inquiry-based mobile learning environment.Frontline Learning Research, 5(4):42–60, 2017

    work page 2017

  11. [11]

    Ryan and Edward L

    Richard M. Ryan and Edward L. Deci. Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being.American Psychologist, 55(1):68–78, 2000

    work page 2000

  12. [12]

    Ann Renninger

    Suzanne Hidi and K. Ann Renninger. The four-phase model of interest development.Educational Psychologist, 41(2):111–127, 2006

    work page 2006

  13. [13]

    Deep interest network for click-through rate prediction

    Guorui Zhou, Xiaoqiang Zhu, Chengru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. Deep interest network for click-through rate prediction. InProceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1059–1068, 2018

    work page 2018

  14. [14]

    E-learning personalization based on hybrid recommendation strategy and learning style identification.Computers & Education, 56(3):885– 899, 2011

    Aleksandra Klašnja-Milicevi´c, Boban Vesin, Mirjana Ivanovi´c, and Zoran Budimac. E-learning personalization based on hybrid recommendation strategy and learning style identification.Computers & Education, 56(3):885– 899, 2011

    work page 2011

  15. [15]

    Multicultural Education Series

    Geneva Gay.Culturally Responsive Teaching: Theory, Research, and Practice. Multicultural Education Series. Teachers College Press, 3rd edition, 2018

    work page 2018

  16. [16]

    Science education and student diversity: Race/ethnicity, language, culture, and socioeconomic status

    Okhee Lee and Aurolyn Luykx. Science education and student diversity: Race/ethnicity, language, culture, and socioeconomic status. In Sandra K. Abell and Norman G. Lederman, editors,Handbook of Research on Science Education, pages 171–197. Lawrence Erlbaum Associates, 1st edition, 2007

    work page 2007

  17. [17]

    Accessed Apr

    Next.js: The full-stack React framework.https://nextjs.org/. Accessed Apr. 2025

    work page 2025

  18. [18]

    Accessed Apr

    AWS Lambda: Serverless computing service.https://aws.amazon.com/lambda/. Accessed Apr. 2025

    work page 2025

  19. [19]

    https://aws.amazon.com/ api-gateway/

    AWS API Gateway: Service for creating and managing APIs at scale. https://aws.amazon.com/ api-gateway/. Accessed Apr. 2025

    work page 2025

  20. [20]

    Accessed Apr

    AWS DynamoDB: Fully managed NoSQL database service.https://aws.amazon.com/dynamodb/. Accessed Apr. 2025

    work page 2025

  21. [21]

    Accessed Apr

    AWS S3: Scalable object storage service.https://aws.amazon.com/s3/. Accessed Apr. 2025

    work page 2025

  22. [22]

    https://langchain.com/

    LangChain: Framework for developing applications powered by language models. https://langchain.com/. Accessed Apr. 2025

    work page 2025

  23. [23]

    https://www.litellm.ai/

    LiteLLM: Unified API for calling LLM providers with load balancing. https://www.litellm.ai/. Accessed Apr. 2025. 10 APREPRINT- MAY8, 2026

    work page 2025

  24. [24]

    https://www.langchain.com/ langsmith-platform

    LangSmith: AI evaluation, debugging, and monitoring platform. https://www.langchain.com/ langsmith-platform. Accessed Apr. 2025

    work page 2025

  25. [25]

    Measuring massive multitask language understanding.Proceedings of the International Conference on Learning Representations (ICLR), 2021

    Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding.Proceedings of the International Conference on Learning Representations (ICLR), 2021

    work page 2021

  26. [26]

    Evaluating Large Language Models Trained on Code

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374, 2021

    work page internal anchor Pith review arXiv 2021

  27. [27]

    https://github.com/openai/simple-evals

    OpenAI simple-evals: Lightweight evaluation library. https://github.com/openai/simple-evals. Ac- cessed Apr. 2025

    work page 2025

  28. [28]

    Model card: Claude 3.5 sonnet

    Anthropic. Model card: Claude 3.5 sonnet. https://www.anthropic.com/news/claude-3-5-sonnet , 2024. Accessed Apr. 2025

    work page 2024

  29. [29]

    Llama 3.3 70B instruct: Model card

    Meta AI. Llama 3.3 70B instruct: Model card. https://huggingface.co/meta-llama/Llama-3. 3-70B-Instruct, 2024. Released Dec. 6, 2024. Accessed Apr. 2025

    work page 2024

  30. [30]

    Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context

    Google DeepMind. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. https://arxiv.org/abs/2403.05530, 2024. Accessed Apr. 2025

    work page internal anchor Pith review arXiv 2024

  31. [31]

    Mistral large 2: A new generation of flagship model

    Mistral AI. Mistral large 2: A new generation of flagship model. https://mistral.ai/news/ mistral-large-2407/, 2024. Accessed Apr. 2025. 11 APREPRINT- MAY8, 2026 Appendix A Prompt Templates We present the full text of a subset of prompts used in Taklif.AI. Variables in braces ({...}) are dynamically populated at runtime by the LangChain pipeline. A.1 Pers...

    work page 2024

  32. [32]

    Customization: Tailoring assignments to suit diverse student abilities

    work page

  33. [33]

    Creativity: Designing engaging and imaginative learning experiences

    work page

  34. [34]

    Cultural Sensitivity: Avoiding stereotypes and incorporating diverse perspectives

    work page

  35. [35]

    Fairness: Ensuring assignments maintain equitable complexity across contexts. ### Objective Transform a general assignment into a personalized, joyful learning experience by integrating the student’s interests while adhering to the specified learning objective. Ensure all information is accurate and grounded in known facts. Be creative only in structuring...

    work page

  36. [36]

    General Assignment (to be personalized): {general_assignment}

    work page

  37. [37]

    Student Interest: {interest} ### Output Requirements

    work page

  38. [38]

    Assignment Title: Keep the title concise and attached with corresponding emojis and icons

    work page

  39. [39]

    - Incorporate elements of the student’s interest to enhance engagement and relevance

    Assignment Content: - Align the assignment with the provided general assignment. - Incorporate elements of the student’s interest to enhance engagement and relevance. - Content length doesn’t exceed a 50% increment in the length of the original assignment length. - Avoid including explicit hints, solutions, or leading questions

    work page

  40. [40]

    - Balance creativity with the rigor required to achieve the learning objective

    Tone and Style: - Ensure the assignment is clear, inclusive, and free from culturally insensitive language. - Balance creativity with the rigor required to achieve the learning objective

    work page

  41. [41]

    ‘json but provide it like this {{}} instead):

    Formatting: - Provide the entire response in JSON format for clarity and structure as follows (do not provide it as “‘json but provide it like this {{}} instead): "{{ "assignment_title": "Title", "assignment_content": "assignment content", }}" - Format the assignment_content in Markdown between " (ensure that the formatting including ### headings and new ...

    work page 2026