Benchmarking Large Language Models for Data Pipeline Code Generation and Execution

Benchmarking Large Language Models for Data Pipeline Code Generation and Execution | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Benchmarking Large Language Models for Data Pipeline Code Generation and Execution Chiara Rucco, Motaz Saad, Tobia Martina, Antonella Longo This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6786102/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract In today’s data-driven landscape, organizations face mounting pressure to accelerate data processing and analysis while minimizing manual engineering efforts. This paper investigates the application of Large Language Models (LLMs) to automate the creation of data pipelines, evaluating their efficacy across code-based (Apache Airflow), low-code (Azure Data Factory), and hybrid (Databricks) platforms. Through systematic experimentation, we demonstrate that LLMs like GPT-4o, Qwen 2.5-Max 72B, and DeepSeek-V3 37B can successfully generate functional pipelines for tasks ranging from basic API interactions to multi-step ETL processes, reducing development time by an estimated 40–60% in code-centric environments.Our findings reveal stark contrasts in platform adaptability: while LLMs excel in code-based systems —resolving ambiguous requirements through logical constructs like Python operators— they struggle with low-code platforms such as Azure Data Factory, where implicit configuration dependencies and JSON syntax constraints lead to systemic failures. This highlights a critical trade-off between the flexibility of code-based orchestration and the accessibility of low-code tools.The study further underscores the importance of prompt engineering in guiding LLM outputs, particularly for complex transformations and platform-specific idiosyncrasies. While LLMs show promise in accelerating routine data engineering tasks, challenges persist in ensuring security, governance, and consistency in advanced patterns like error handling. This research contributes to the evolving discourse on AI-augmented data engineering, illustrating how LLMs can transform pipeline development, not by replacing human engineers, but by enhancing productivity and enabling focus on high-value architectural innovation. Code Generation LLM Pipelines ETL Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6786102","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":464424428,"identity":"41504d17-6ab7-4a6a-9c05-8e7fe96b369c","order_by":0,"name":"Chiara Rucco","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABA0lEQVRIiWNgGAWjYDACCcYGBOcDAwMPAwMzRISNGC2MM8BaQCIJQC249EggsZl5IDohWnBZwz+7ufFzRQ1DvsHxww8f27bdkZFvP9j4uPCHDQOffANWLRJ3DjZLnjnGYLnhTJqxcW7bMx6DM4nNxjMS0nA6zEAisUGygY3BwOAGD5t0btthHgOGxDZpnoTD+LQ0/2z4B9ViCdQi3/+w/TcBLW2SjW1QLYxALQw3EtuY8WmRACqwbOyTMJAE+sWw5xzQYTceNkvzpKXxsLElYA+xGemPbzZ8szHgA4bYgx9lh+3l+5MPfuaxsZGTbz6A3RqoZZhCPPjUj4JRMApGwSjADwAhhlQ1sycZXwAAAABJRU5ErkJggg==","orcid":"","institution":"University of Salento","correspondingAuthor":true,"prefix":"","firstName":"Chiara","middleName":"","lastName":"Rucco","suffix":""},{"id":464424429,"identity":"af534ad6-13fa-4aec-8da3-c91da39060f1","order_by":1,"name":"Motaz Saad","email":"","orcid":"","institution":"University of Salento","correspondingAuthor":false,"prefix":"","firstName":"Motaz","middleName":"","lastName":"Saad","suffix":""},{"id":464424430,"identity":"857f2d09-7370-4bf4-a7da-8e9d111e308a","order_by":2,"name":"Tobia Martina","email":"","orcid":"","institution":"University of Salento","correspondingAuthor":false,"prefix":"","firstName":"Tobia","middleName":"","lastName":"Martina","suffix":""},{"id":464424431,"identity":"effcaedc-d92a-4ab2-b51d-628b47aa5f31","order_by":3,"name":"Antonella Longo","email":"","orcid":"","institution":"University of Salento","correspondingAuthor":false,"prefix":"","firstName":"Antonella","middleName":"","lastName":"Longo","suffix":""}],"badges":[],"createdAt":"2025-05-30 15:38:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6786102/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6786102/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93059303,"identity":"ce1b4368-e48b-4da8-9114-1a79b75daf92","added_by":"auto","created_at":"2025-10-08 15:32:18","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":730697,"visible":true,"origin":"","legend":"","description":"","filename":"DataPipelinegenerationusingLLMv21.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6786102/v1_covered_bfd57749-b171-4292-8dbd-ef892967e641.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Benchmarking Large Language Models for Data Pipeline Code Generation and Execution","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Code Generation, LLM, Pipelines, ETL","lastPublishedDoi":"10.21203/rs.3.rs-6786102/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6786102/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn today\u0026rsquo;s data-driven landscape, organizations face mounting pressure to accelerate data processing and analysis while minimizing manual engineering efforts. This paper investigates the application of Large Language Models (LLMs) to automate the creation of data pipelines, evaluating their efficacy across code-based (Apache Airflow), low-code (Azure Data Factory), and hybrid (Databricks) platforms. Through systematic experimentation, we demonstrate that LLMs like GPT-4o, Qwen 2.5-Max 72B, and DeepSeek-V3 37B can successfully generate functional pipelines for tasks ranging from basic API interactions to multi-step ETL processes, reducing development time by an estimated 40\u0026ndash;60% in code-centric environments.Our findings reveal stark contrasts in platform adaptability: while LLMs excel in code-based systems \u0026mdash;resolving ambiguous requirements through logical constructs like Python operators\u0026mdash; they struggle with low-code platforms such as Azure Data Factory, where implicit configuration dependencies and JSON syntax constraints lead to systemic failures. This highlights a critical trade-off between the flexibility of code-based orchestration and the accessibility of low-code tools.The study further underscores the importance of prompt engineering in guiding LLM outputs, particularly for complex transformations and platform-specific idiosyncrasies. While LLMs show promise in accelerating routine data engineering tasks, challenges persist in ensuring security, governance, and consistency in advanced patterns like error handling. This research contributes to the evolving discourse on AI-augmented data engineering, illustrating how LLMs can transform pipeline development, not by replacing human engineers, but by enhancing productivity and enabling focus on high-value architectural innovation.\u003c/p\u003e","manuscriptTitle":"Benchmarking Large Language Models for Data Pipeline Code Generation and Execution","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-02 07:40:08","doi":"10.21203/rs.3.rs-6786102/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f5540286-5255-4a03-8d13-d66b95f184fa","owner":[],"postedDate":"July 2nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-10-08T15:23:59+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-02 07:40:08","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6786102","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6786102","identity":"rs-6786102","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}