{"resourceId":"frontiers-feng-university-pe-llm-rag-2026-09-08","versions":[{"version":"external-36f9879e249b6709aac1ab43a1c1dd82fd26e1516babbde3a77af4cb2cd4faea","resource":{"id":"frontiers-feng-university-pe-llm-rag-2026-09-08","title":"Frontiers: university PE LLM+RAG platform improves need satisfaction but not reported activity","organization":"Tianjin University of Technology and collaborating Chinese universities","sector":"Higher education physical education / campus sports platforms","geography":"China (university PE; transfer limits to U.S. NCAA athletics)","publishedAt":"08 September 2026","publicationDate":"2026-09-08","eventDate":null,"sourceName":"Frontiers in Public Health","sourceLabel":"Peer-reviewed academic research","sourceUrl":"https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2026.1835590/full","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["knowledge-work","developers-agents","data-security","accessibility-workforce","operating-model"],"finding":"Feng et al. (published 2026-09-08) describe a university digital-intelligent sports platform combining LLM + RAG, institutional knowledge resources, and campus data, with a formative deployment (428 students, six instructors) and an independent six-week non-randomized intact-class quasi-experiment (n=196). Platform-supported PE showed higher adjusted need satisfaction (difference 0.618, 95% CI 0.506–0.730, p<0.001) and lower need frustration (−0.653, 95% CI −0.759 to −0.547, p<0.001); self-reported physical activity difference was not significant (p=0.216). Authors stress teacher review, non-randomization, response-pattern concentration, and no claim of fitness-volume gains.","sledRelevance":"Measured higher-ed sports AI evidence for LLM+RAG teaching support and psychological-need outcomes—with explicit limits—useful as transfer-with-caution for campus PE/athlete-support tooling, not as U.S. varsity competitive-performance proof.","evidence":"Peer-reviewed methods, tables for baseline-adjusted outcomes, ethics approval ID TUT-R-202603270029, and limitations section detailing intact-class design and self-report measures.","architectureImplications":"LLM + RAG over institutional PE corpora, role-based campus data, human-in-the-loop corrections, and externally hosted DeepSeek-R1 in the reported deployment—emphasizing retrieval grounding over unconstrained generation.","governanceImplications":"Teacher verification of lesson plans/guidance, de-identification before analysis, and staged evaluation before agent-style tool use are treated as required controls.","securityPrivacyImplications":"Layered access, data minimization, and de-identification are described; federated learning/differential privacy are prospective. Cross-border transfer to U.S. NCAA contexts needs local FERPA/HIPAA-aligned redesign.","caveats":"Chinese university PE (not U.S. NCAA varsity athletics); non-randomized four-class design; self-report outcomes with patterned responses more concentrated in the platform arm; activity volume unchanged over six weeks.","streamIds":["college-athletics"],"roles":{"sales":"Campus recreation and academic PE leaders want AI tutors that personalize guidance without replacing instructors. Discovery should ask whether the use case is required PE courses versus varsity performance, what knowledge corpus is licensed, and how teachers edit model output. Bound value is improved autonomy/competence/relatedness experiences in PE classes—not proven injury reduction or varsity win rates. Unsupported claims include U.S. Power-conference transfer, guaranteed MET-minute increases, or medical exercise prescription.","engineering":"Fit is RAG-grounded teaching/exercise guidance with instructor review, not autonomous return-to-play. Integration needs LMS/IdP, knowledge-base pipelines, and optional wearable feeds under minimization. Prerequisites include curated PE corpus, teacher review workflow, and ethics approval for student data. Constraints include external model hosting and non-causal classroom design. Security must separate identifiers from analytics and block high-stakes medical advice. Proof of value: factuality/teacher-correction rates plus independent psychometrics—not only need-satisfaction deltas.","delivery":"Delivery includes corpus curation, instructor training, classroom pilot with control comparison, and response-quality monitoring. Dependencies: PE faculty buy-in, campus identity, and licensed content. Ownership: PE department for pedagogy; IT for platform ops. Skills adoption centers on teachers editing AI drafts and students reporting bad answers. Governance checkpoints: pre-publish teacher approval, monthly unsafe-output review, and stop rules for medical claims. Acceptance criteria: documented teacher-edit rate, no significant adverse events, and no claim of activity-volume gains without measures. Risks include overclaiming causal effects and exporting China-specific stacks unchanged."},"roleDetails":{"sales":{"customerProblem":"University PE programs struggle to personalize instruction and motivation at scale with limited teaching staff.","stakeholders":["PE faculty / department chair","Campus IT / teaching innovation","Students in required PE courses","Institutional research / ethics board"],"discoveryQuestions":["Is the target required PE, recreation, or varsity athletics?","What corpus and model hosting constraints exist?","How will success be measured beyond self-report surveys?"],"valueHypothesis":"RAG-grounded PE assistants can improve students’ psychological-need experiences when teachers remain accountable for content.","potentialEngagement":"One-term PE pilot with pre/post need measures, teacher-correction logging, and explicit transfer assessment for any athletics use.","unsupportedClaims":["Direct NCAA competitive-performance gains","Guaranteed increases in physical activity volume","Medical or injury diagnosis capability"]},"engineering":{"fit":"LLM+RAG teaching and exercise-guidance copilot for university PE with human review—not varsity injury AI.","architectureIntegration":"Retrieve institutional PE materials; generate revisable lesson/guidance drafts; log feedback; keep high-stakes decisions with instructors.","prerequisites":["Licensed/owned knowledge corpus","Instructor review UI","AuthN and de-identification pipeline"],"constraints":["Non-randomized evidence base","External LLM hosting in reported deployment","Self-report outcome limits"],"security":"Minimize student identifiers, restrict wearable/health fields, and prevent model training on sensitive campus records without approval.","proofOfValue":["Teacher-correction and unsupported-claim rates","Blind review of sample outputs","Replication with more classes / objective activity measures"]},"delivery":{"work":"Deploy RAG corpus, train instructors, run controlled classroom pilot, and monitor response quality.","dependencies":["Faculty participation","Campus IdP/LMS","Ethics approval"],"ownership":"PE department owns pedagogy; IT owns hosting and access control.","skillsAdoption":"Instructors practice editing AI drafts and teaching students to challenge outputs.","governanceCheckpoints":["Mandatory teacher approval for plans/guidance","Monthly safety/factuality sampling","Scope review before any varsity-athlete expansion"],"acceptanceCriteria":["Need-experience metrics pre-registered and reported with limitations","Zero unreviewed medical advice in production","Written transfer memo before NCAA athletics use"],"risks":["Causal overclaim from intact-class design","Privacy gaps if wearables expand","Misapplication to competitive performance decisions"]}},"retrievedAt":"2026-09-28T03:17:20Z","enrichedAt":"2026-09-28T03:28:00Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Targets PE teaching load with AI lesson prep and guidance while insisting teachers retain instructional decisions; student accessibility depends on mobile access and clear language.","procurementImplications":"Prefer platforms that expose retrieval sources, teacher-edit workflows, and local hosting options over opaque PE chatbots claiming fitness gains.","operatingModelImplications":"Shifts PE toward AI-assisted, teacher-mediated personalization; varsity athletic departments should treat transfer as a separate operating decision.","sourceVerification":{"openedUrl":"https://www.frontiersin.org/journals/public-health/articles/10.3389/fpubh.2026.1835590/full","referenceExcerpt":"Baseline-adjusted analyses indicated higher posttest basic psychological need satisfaction in the platform-supported group","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}