Multi-agent generate–validate–revise framework with four specialized validator agents for math problem personalization

Design · All designs
Evidence · 2 claims (1 for, 1 mixed) · 1 study (1 design), q2 · 0 of 1 report an effect size · 2 claims rest on one study

Description

The framework formalizes context personalization of math problems as an iterative generate–validate–revise loop. A conversion agent generates an initial topic-aligned candidate problem, and "four independentvalidator agentsthat return a binary decision (pass/fail) and brief diagnostic feedback" check the candidate against solvability, realism, readability, and authenticity criteria; one or more refinement agents then revise the candidate using validator feedback until all validators pass or a maximum iteration count is reached. The framework is evaluated on 600 ASSISTments problems personalized to 20 student interest topics.

Design Implications

Context

Requirements

• A target topic reflecting a learner's interest, a target grade reading level computed with the Flesch–Kincaid metric, and validator feedback to drive refinement

Constraints

• Refinement iterates for at most k iterations (set to k=3 in the evaluation) and stops early only if the candidate passes all validator agent checks

Target Learners

• middle school mathematics students

Learning Goals

• math word problem solving with personally interesting contexts

Claims

• Single Refinement Iteration Reduces Personalization Failures [+M]
• Validator Reliability Highest Realism Lowest Authenticity [~M]

Related Designs

• Four specialized evaluation agents for LLM-generated problems
• Multi-agent teacher-in-the-loop architecture for context personalization

Examples

-

Key Sources

• Fareya Ikram, Nischal Ashok Kumar, Junyang Lu, Hunter McNichols, Candace Walkington, Neil Heffernan, and Andrew S. Lan. (2026). A Multi-Agent Approach to Validate and Refine LLM-Generated Personalized Math Problems. arXiv preprint. https://arxiv.org/abs/2604.05160

Page metadata

Field	Value
type	design
id	multi-agent-generate-validate-revise-personalization-framework
title	Multi-agent generate–validate–revise framework with four specialized validator agents for math problem personalization
description	The framework formalizes context personalization of math problems as an iterative generate–validate–revise loop.
status	review
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content
sources	1 — fareya-ikram-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.