Dimension-Direct Routing: Achieving 25% Depth Improvement in Multi- Model LLM Systems via Explicit Capability Factorization

Dimension-Direct Routing: Achieving 25% Depth Improvement in Multi- Model LLM Systems via Explicit Capability Factorization | 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 Dimension-Direct Routing: Achieving 25% Depth Improvement in Multi- Model LLM Systems via Explicit Capability Factorization Tao Rui This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9317311/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 Large Language Models (LLMs) exhibit distinct capabilities across different knowledge domains, yet single-model deployments struggle with knowledge-intensive tasks requiring cross-domain reasoning. We present eVoiceClaw Desktop, a multi-model orchestration system that operationalizes an \"AI managing AI\" paradigm: instead of humans manually selecting models, the system dynamically routes complex queries to specialized models through a dimension-direct routing algorithm.\n\nThe system underwent four major configuration iterations (V1–V4), culminating in V5 that addresses critical challenges in cross-domain task allocation and semantic accumulation bias. V5 achieves a 98% workflow trigger rate across 50 benchmark questions in Chinese, leveraging 12 models with balanced diversity (top model ≤16% usage share).\n\nWe evaluate response quality using LLM-as-Judge (Claude Opus 4.6) across four dimensions: factual accuracy, completeness, depth, and structure. Compared to single-model baselines, V5 achieves a 14.3% overall quality improvement, with depth of analysis improving by 25.9%, at the expense of approximately 9× higher latency and cost.\n\nAs a meta-demonstration, the initial draft of this paper was itself generated by the system (see Appendix B). Artificial Intelligence and Machine Learning LLM routing multi-model orchestration dimension-direct routing capability factorization model selection LLM-as-Judge mixture of experts semantic accumulation bias knowledge-intensive tasks AI managing AI Full Text Additional Declarations The authors declare no competing interests. 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-9317311","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":617391317,"identity":"bf3b937e-0ca2-406e-92ca-9bb38b5839a7","order_by":0,"name":"Tao Rui","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAy0lEQVRIiWNgGAWjYHACAzDJxt4AYluQooXnAIgtQYIWBokEMElYvXl787YHH9vuJfZJPr+64UeBBAN/e3cCXi0yZ46VG85sK05sk84pu9kDdJjEmbMb8GqRkMgxk+ZtSwBpSbvBA9RiIJFLhJa/IC2SZ9Ju/iFaCyNIiwT7sdvE2cJzrEyy51yCcRtPDtttGQMJHsJ+YW/eJvGjLEF2fvvxZzff/LGR42/vxa8FDBjZQCQPOIJ4CCsHgz8ggv0BkapHwSgYBaNgpAEAAWFDDEk4UskAAAAASUVORK5CYII=","orcid":"https://orcid.org/0009-0004-3898-4655","institution":"Independent Researcher","correspondingAuthor":true,"prefix":"","firstName":"Tao","middleName":"","lastName":"Rui","suffix":""}],"badges":[],"createdAt":"2026-04-04 04:06:42","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9317311/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9317311/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106404397,"identity":"a8c8f2ee-1d81-45b6-9b5d-38141c355242","added_by":"auto","created_at":"2026-04-08 09:15:56","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2319398,"visible":true,"origin":"","legend":"","description":"","filename":"eVoiceClawDesktopPaper20260402.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9317311/v1_covered_db971055-1c91-4dec-824f-293b9abc399e.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eDimension-Direct Routing: Achieving 25% Depth Improvement in Multi- Model LLM Systems via Explicit Capability Factorization\u003c/p\u003e","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Independent Researcher","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"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":"LLM routing, multi-model orchestration, dimension-direct routing, capability factorization, model selection, LLM-as-Judge, mixture of experts, semantic accumulation bias, knowledge-intensive tasks, AI managing AI","lastPublishedDoi":"10.21203/rs.3.rs-9317311/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9317311/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLarge Language Models (LLMs) exhibit distinct capabilities across different knowledge domains, yet single-model deployments struggle with knowledge-intensive tasks requiring cross-domain reasoning. We present eVoiceClaw Desktop, a multi-model orchestration system that operationalizes an \\\"AI managing AI\\\" paradigm: instead of humans manually selecting models, the system dynamically routes complex queries to specialized models through a dimension-direct routing algorithm.\\n\\nThe system underwent four major configuration iterations (V1–V4), culminating in V5 that addresses critical challenges in cross-domain task allocation and semantic accumulation bias. V5 achieves a 98% workflow trigger rate across 50 benchmark questions in Chinese, leveraging 12 models with balanced diversity (top model ≤16% usage share).\\n\\nWe evaluate response quality using LLM-as-Judge (Claude Opus 4.6) across four dimensions: factual accuracy, completeness, depth, and structure. Compared to single-model baselines, V5 achieves a 14.3% overall quality improvement, with depth of analysis improving by 25.9%, at the expense of approximately 9× higher latency and cost.\\n\\nAs a meta-demonstration, the initial draft of this paper was itself generated by the system (see Appendix B).\u003c/p\u003e","manuscriptTitle":"Dimension-Direct Routing: Achieving 25% Depth Improvement in Multi- Model LLM Systems via Explicit Capability Factorization","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-07 19:18:58","doi":"10.21203/rs.3.rs-9317311/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":"49861f91-2e11-4601-a45a-aaf9ce8c6eb5","owner":[],"postedDate":"April 7th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":65705734,"name":"Artificial Intelligence and Machine Learning"}],"tags":[],"updatedAt":"2026-04-07T19:18:58+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-07 19:18:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9317311","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9317311","identity":"rs-9317311","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}