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

Synthesis: Kurdya and colleagues (2026) present Taklif.AI, a platform that uses large language models to generate personalized college assignments from students' own interests and cultural contexts rather than from academic-performance metrics alone. The stated problem is educators' difficulty in producing engaging assignments that fit diverse interests and cognitive abilities: traditional one-size-fits-all tasks are linked in the paper to reduced engagement and increased reliance on unethical practices such as plagiarism. The platform uses a structured Prompt Engineering pipeline with input and output Guardrails, a serverless AWS architecture with a Next.js frontend, Llama 3.3 70B as the primary model via LiteLLM for multi-provider load balancing, and LangChain for prompt orchestration. Preliminary user-acceptance testing with 68 participants (65 students and three educators) found positive reception, with 84% rating the personalization feature as beneficial; the authors treat rigorous empirical evaluation of learning outcomes as future work.

Taklif.AI differs from performance-driven personalization by incorporating students' extracurricular interests and cultural contexts into assignment generation — an approach aligned with Culturally Relevant Pedagogy principles and with the wider Personalized Learning literature on interest-based task design.

System architecture highlights:

• Primary LLM: Llama 3.3 70B via LiteLLM for multi-provider load balancing
• Prompt orchestration: LangChain with structured prompt engineering pipeline
• Quality control: Input and output Guardrails to ensure assignment quality
• Infrastructure: AWS serverless with Next.js frontend

User testing results (n=68):

• 65 students and 3 educators participated
• 84% rated the personalization feature as beneficial
• Strong user acceptance overall

The platform represents a shift from one-size-fits-all assignments toward interest-driven engagement, potentially reducing Academic Integrity concerns like plagiarism. However, the paper acknowledges the need for rigorous empirical evaluation of learning outcomes beyond user acceptance — echoing the Distinguishing performance gains from learning when using generative AI distinction between engagement gains and actual learning improvements.

This work connects to the broader Automated Question Generation literature and the Edtech Platform ecosystem. The use of open-weight models (Llama 3.3 70B) aligns with the trend toward institution-controlled AI deployment seen in LaTA: A Drop-in, FERPA-Compliant Local-LLM Autograder for Upper-Division STEM Coursework and From Surface Learning to Deep Understanding: A Grounded AI Tutoring System for Moodle.

What this means for practice

• Instructors. Collect students' extracurricular interests and cultural contexts at the start of a term and use them as the context for personalized assignments: in the platform's user acceptance test, 84% of participants (57/68) rated the personalization feature as beneficial and 91% (62/68) rated the interface as intuitive.
• Instructors. Cap how far a personalized rewrite may drift from the original task — Taklif.AI's prompt restricts generated content to a 50% increment over the length of the source assignment, which keeps interest-based framing a change of context rather than of learning objectives.
• Instructors. Grade the output yourself against a shared objective-based rubric: the platform does not evaluate or assess student submissions, and each personalized assignment produces a different problem statement, which complicates cross-student comparison.
• Designers. Build input and output Guardrails before scaling up. Against approximately 200 development test inputs, the interest guardrails rejected about 12% of submitted interests, the assignment guardrails rejected about 8% of input documents, and the output guardrails flagged about 3% of generated assignments for out-of-scope or potentially hallucinated content.
• Designers. Budget for generation latency inside the class period: under a simulated load of 100 concurrent users generating assignments, end-to-end response time averaged 30 seconds per request.

Limitations

• No controlled experiment: the user acceptance test measured perceived benefit only, and the paper reports no comparison of grades, comprehension, or plagiarism rates against standard or manually personalized assignments.
• Small, single-site sample: all 68 participants (65 undergraduate students and 3 educators) came from the Faculty of Information Technology at the Islamic University of Gaza, so the results may not generalize to other institutions or disciplines.
• Self-report instrument: four 5-point Likert items plus one open-ended question, administered in sessions of about 12 minutes each, capture perceived engagement and ease of use rather than learning gains.
• The guardrail rates (12%, 8%, 3%) come from development-phase testing with approximately 200 test inputs and were not validated against labeled ground truth; the authors also report no systematic human evaluation of output quality and no comparison baselines.

Citation

Kurdya, Z., Zuqlam, M., Amassi, S., Telbany, S., & Saad, M. (2026). Taklif.AI: LLM-powered platform for interest-based personalized college assignments.