[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85740-en":3,"doc-seo-85740-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},85740,5909877438554,"Maeve","https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272",8,"Research & Report","Exploring Agentic Workflows for Generating High Quality Math Visual Aids","Mathematical diagrams are central to K-12 learning as both problem components and scaffolding for understanding, yet existing AI tools struggle to generate visuals that are reliably accurate and pedagogically appropriate from textual descriptions. State-of-the-art LLMs reach only a 73.9% success rate on middle-school math diagram generation, leaving a substantial quality gap. This work introduces an agentic workflow where LLM agents generate diagram QA questions and VLMs evaluate and iteratively refine outputs in real time, improving reliability and educational value.","Exploring Agentic Workflows for Generating High Quality Math Visual Aids  \nRizwaan Malik  \nStanford Graduate School of Education [rizmalik@stanford.edu](rizmalik@stanford.edu)  \nAshna Khetan  \nDepartment of Computer Science  \n[ashnak@stanford.edu](ashnak@stanford.edu)  \nIsabel Sieh  \nDepartment of Computer Science  \n[isabelrs@stanford.edu](isabelrs@stanford.edu)  \narXiv :2607 .09839v 1 [ cs .AI] 10 Jul 2026  \nSamin Khan  \nStanford Graduate School of Education  \n[samink@stanford.edu](samink@stanford.edu)  \nAbstract  \nMathematical diagrams play a crucial role in K-12 education, both as problem components and as scaffolding for student comprehension [1] . However, current AI tools, including Large Language Models (LLMs), struggle to reliably generate accurate and pedagogically sound visual diagrams, even when provided with detailed descriptions [3, 4] . While state-of-the-art LLMs achieve a 73.9% success rate in generating diagrams for middle-school math problems [5], a significant gap remains. To address this, we introduce anagentic workflow that empowers LLM agents to evaluate the quality of generated visuals and use this feedback to dynamically improve their outputsin real time. This self-improvement loop aims to enhance the accuracy and educational appropriateness of AI-generated diagrams. Our research investigates the following questions: First, can LLMs accurately generate quality assurance (QA) questions for a visual aid, given specific criteria for visual aid quality? Second, given valid QA questions, can Vision Language Models (VLMs) effectively vet generated K-12 visual aids and use this to iteratively improve outputs? We conduct an exploratory evaluation of our agentic workflow. We identify key areas for improvement, including enhancing spatial reasoning and ensuring comprehensive coverage of diagram features in QA questions. Our results provide preliminary evidence that this approach can improve the reliability and educational value of AI-generated mathematical diagrams.  \n1 Introduction  \nMathematical diagrams are fundamental to K-12 education, serving both as essential components of problems and as scaffolding tools for student comprehension [1] . Diagrams help to visualize abstract concepts, making them more accessible and understandable for students of all learning styles. For example, diagrams can clarify geometric relationships, illustrate proportional reasoning, and support problem-solving strategies. The ability to effectively create and utilize mathematical diagrams is therefore a crucial skill for both students and  \nPreprint. Under review.  \neducators. However, manually creating these diagrams can be time-consuming and require a certain level of technical expertise.  \nThe advent of Artificial Intelligence (AI) offers a promising opportunity to automate the generation of mathematical diagrams, potentially freeing up educators’ time and providing students with personalized learning experiences. AI tools show promise in supporting teachers with lesson preparation; however, they still struggle with reliably generating visual diagrams [6, 7] . This is particularly true for diagrams that require precise geometric relationships or symbolic representations. Furthermore, generated diagrams must not only be accurate but also pedagogically sound, meaning they should effectively convey the intended mathematical concepts and be appropriate for the target student population.  \nWhile progress has been made in using Large Language Models (LLMs) to generate code or descriptions for mathematical diagrams, the accuracy of these generated visuals remains a significant challenge. When provided with specific, detailed diagram descriptions, stateof-the-art LLMs achieve a 73.9% success rate within the middle-school domain [5] . This indicates that even with precise instructions, current LLMs struggle to consistently generate diagrams that meet the required standards of accuracy and pedagogical effectiveness. Closing this gap is es","cbCaibVJ0DQvzYUs","https://ap.wps.com/l/cbCaibVJ0DQvzYUs","pdf",589396,1,13,"English","en",105,"# Abstract\n# Introduction\n## Mathematical Visualization Generation\n# Related Work","[{\"question\":\"What problem does the document address in AI-generated math diagrams?\",\"answer\":\"It addresses the difficulty of producing diagrams that are both accurate and pedagogically sound for K-12 education, even when detailed descriptions are provided to LLMs.\"},{\"question\":\"How does the proposed agentic workflow improve diagram quality?\",\"answer\":\"An LLM agent generates quality assurance (QA) questions for a visual, and a vision language model (VLM) uses them to evaluate the diagram, then applies the feedback to iteratively refine outputs in real time.\"},{\"question\":\"What research questions does the paper investigate?\",\"answer\":\"It asks whether LLMs can generate effective QA questions given criteria for diagram quality, and whether VLMs can vet K-12 visual aids using those QA questions to improve outputs iteratively.\"}]",1784205945,33,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"exploring-agentic-workflows-for-generating-high-quality-math-visual-aids","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/exploring-agentic-workflows-for-generating-high-quality-math-visual-aids/85740/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does the document address in AI-generated math diagrams?","Question",{"text":75,"@type":76},"It addresses the difficulty of producing diagrams that are both accurate and pedagogically sound for K-12 education, even when detailed descriptions are provided to LLMs.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the proposed agentic workflow improve diagram quality?",{"text":80,"@type":76},"An LLM agent generates quality assurance (QA) questions for a visual, and a vision language model (VLM) uses them to evaluate the diagram, then applies the feedback to iteratively refine outputs in real time.",{"name":82,"@type":73,"acceptedAnswer":83},"What research questions does the paper investigate?",{"text":84,"@type":76},"It asks whether LLMs can generate effective QA questions given criteria for diagram quality, and whether VLMs can vet K-12 visual aids using those QA questions to improve outputs iteratively.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]