Advisor-in-the-Loop, FERPA-First Learning Analytics for Early Student-Success Intervention: Design and Prototype Evaluation of the Inner Compass Platform

Undergraduate attrition remains a persistent challenge for U.S. higher education: roughly one in four entering students does not persist into a second year of study, and advising staff typically carry caseloads far too large for consistent early intervention. This paper presents the design rationale and prototype evaluation of Inner Compass, a student-success analytics platform built around two constraints that are usually treated as afterthoughts: (1) an advisor-in-the-loop operating model, in which the system never contacts students and never automates an academic decision, and (2) a FERPA-first data architecture that operates on data institutions already collect and routes every output to a named professional. The minimum-viable prototype ingests learning-management-system (LMS) activity and student-information-system (SIS) records, and produces three advisor-facing outputs: an engagement-pattern profile, a calibrated non-retention risk flag with per-student drivers, and a program pathway-fit signal. On a held-out split of a synthetic labelled cohort (n = 188), the calibrated risk classifier attains 67.6% accuracy and ROC-AUC of 0.72, with an operating point deliberately tuned toward recall on the at-risk class (76.2% at-risk recall). We map each architectural commitment to the trustworthiness characteristics of the NIST AI Risk Management Framework, describe a staged data-governance protocol for deployment, and specify a pilot design for a single academic program. All reported metrics are produced by the prototype on synthetic demonstration data and are intended to demonstrate end-to-end system function, not outcomes at any institution. We argue that this advisor-mediated, compliance-first configuration is the design position most consistent with current federal guidance on AI in education, and with how advising offices actually operate.