{"resourceId":"quincoso-aurora-neurosymbolic-advising-sac2026","versions":[{"version":"external-1564d7fdec9f9da695f0dea4c4cb1772306ece8f1bf15f99ab8c0b12b93e167b","resource":{"id":"quincoso-aurora-neurosymbolic-advising-sac2026","title":"SAC 2026: Aurora neuro-symbolic advising agent lifts semantic alignment to 0.93 vs raw LLM 0.68","organization":"Florida International University / Northeastern University (Quincoso Lugones et al.)","sector":"Higher education academic advising / AI systems","geography":"United States (FIU-centered advising research; SAC ’26 Thessaloniki publication)","publishedAt":"2026-03-23","publicationDate":"2026-03-23","eventDate":"2026-03-23","sourceName":"ACM SAC 2026","sourceLabel":"Peer-reviewed systems paper on neuro-symbolic degree advising","sourceUrl":"https://dl.acm.org/doi/pdf/10.1145/3748522.3779850","evidenceClass":"academic-research","outcomeClass":"emerging","topics":["knowledge-work","developers-agents","data-security","operating-model","governance-procurement"],"finding":"Aurora combines BCNF curricular databases, a Prolog engine for prerequisites/credits, and an instruction-tuned LLM to deliver policy-compliant advising explanations. On a structured evaluation suite spanning common and edge advising scenarios, Aurora improved semantic alignment with expert-crafted answers from 0.68 (raw LLM baseline) to 0.93 (+36%) and achieves perfect precision and recall on nearly half of in-scope cases, with correct fallbacks for out-of-scope prompts. Runtime averaged about 33 seconds per query versus ~20 for the raw LLM. Authors motivate the work with advisor-to-student ratios commonly exceeding 300:1. This is recommendation-quality evidence, not a student persistence RCT.","sledRelevance":"U.S. public-university advising architecture signal: grounding GenAI in symbolic degree rules can sharply raise policy alignment versus raw LLMs—relevant as campuses deploy AI advising under FERPA and audit pressure.","evidence":"ACM SAC ’26 paper DOI 10.1145/3748522.3779850; structured scenario evaluation with cosine/precision-recall metrics; FIU author team.","architectureImplications":"Separate catalog truth (normalized DB + Prolog) from NLG explanation; refuse or escalate out-of-scope advising; expect higher latency than raw LLM chat.","governanceImplications":"Advisors remain accountable for degree decisions; log recommendation trails for appeals; do not auto-execute registration from LLM text alone.","securityPrivacyImplications":"Degree plans and holds are education records—keep engines and logs under institutional control; minimize exporting identifiable plans to consumer LLMs.","caveats":"Scenario/expert-alignment evaluation, not a field RCT on retention or credit completion; ~33s latency may hinder walk-in advising UX; catalog maintenance is an operational dependency.","streamIds":["student-success"],"roles":{"sales":"For advising leaders facing 300:1 caseloads and unsafe raw-LLM degree answers, Quincoso Lugones et al.’s SAC 2026 Aurora system shows a neuro-symbolic stack lifting semantic alignment with expert answers from 0.68 to 0.93 (+36%) with strong precision/recall on many in-scope cases. Sell policy-grounded advising copilots with symbolic prerequisite engines and explicit out-of-scope fallbacks—not chatbots that freestyle the catalog. Discovery must ask who maintains the degree rules database and how advisors override recommendations.","engineering":"Engineering fit is RAG-plus-Prolog (or equivalent rules) over a normalized catalog, with LLM natural-language explanation and auditable trails. Prerequisites include clean program requirement data and identity-aware student context. Constraints: evaluation is scenario-based; latency ~33s; catalog drift breaks guarantees. Security: treat plans/holds as FERPA records; keep inference in university tenancy; escalate rather than guess on ambiguous policy.","delivery":"Delivery is an advising-copilot pilot in one college: load catalog rules, validate on edge cases with lead advisors, set escalation SLAs, and measure recommendation accuracy and advisor time—not only chatbot deflection. Advising leadership owns policy; registrar owns catalog truth; academic technology owns the platform. Acceptance criteria include documented precision on in-scope queries and safe refusal behavior. Risks include over-trust in 0.93 similarity and neglecting field outcomes."},"roleDetails":{"sales":{"customerProblem":"Advisors cannot scale policy-accurate guidance; raw LLMs hallucinate requirements and create compliance risk.","stakeholders":["advising leadership","registrar","academic technology","student success","compliance"],"discoveryQuestions":["Is the degree catalog available in structured, maintained form for rules encoding?","Which advising scenarios must refuse or escalate rather than answer?","What audit trail is required for appeals and FERPA access requests?"],"valueHypothesis":"Neuro-symbolic grounding can raise advising answer alignment with expert policy versus raw LLMs, enabling safer AI assistance at high caseloads.","potentialEngagement":"College-level advising-copilot pilot with rules encoding, scenario validation, and advisor-in-the-loop release.","unsupportedClaims":["Proven retention or graduation lifts from Aurora-style systems","That 0.93 cosine similarity equals perfect catalog compliance in production","Replacement of human advisors"]},"engineering":{"fit":"Fit where program rules can be encoded and advisors will review AI drafts.","architectureIntegration":"BCNF/catalog DB → symbolic checker → RAG context → LLM explanation → advisor UI with fallback states.","prerequisites":["Maintained curriculum data","Advisor workflow integration","Out-of-scope taxonomy"],"constraints":["Scenario evaluation not field RCT","Higher latency than raw chat","Ongoing rules engineering labor"],"security":"FERPA-aligned storage of plans and chats; role-based advisor access; no consumer LLM exfiltration of identifiable plans.","proofOfValue":["Expert-alignment and precision/recall on held-out scenarios","Correct refusal rate on out-of-scope prompts","Advisor time-to-accurate-answer in pilot clinics"]},"delivery":{"work":"Encode one college’s requirements, validate edge cases, deploy advisor-facing copilot with escalation, and review accuracy weekly.","dependencies":["Registrar data quality","Advisor champions","Platform hosting"],"ownership":"Advising dean/director with registrar co-owner for catalog truth.","skillsAdoption":"Train advisors to verify symbolic checks and treat LLM prose as draft explanation.","governanceCheckpoints":["Pre-launch policy sign-off","Monthly hallucination/escalation review","Term outcome check before expansion"],"acceptanceCriteria":["Documented in-scope accuracy targets met in pilot","Fallback path live","No unsupervised auto-registration"],"risks":["Catalog drift causing silent errors","Latency harming adoption","Students treating AI plans as official without advisor confirmation"]}},"retrievedAt":"2026-09-26T03:16:20Z","enrichedAt":"2026-09-26T03:21:39Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Policy-accurate AI drafts may free advisors for complex/disability/basic-needs cases; ensure student-facing outputs meet accessibility standards and that human appointments remain available.","procurementImplications":"Prefer advising AI with inspectable rules engines and audit logs over opaque generative chat; budget for catalog engineering, not only LLM tokens.","operatingModelImplications":"Operating model assigns registrar ownership of truth data and advising ownership of release decisions.","sourceVerification":{"openedUrl":"https://dl.acm.org/doi/pdf/10.1145/3748522.3779850","referenceExcerpt":"from 0.68 to 0.93 (+36%) and achieves perfect precision and recall on","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}