JMIR Publications

ABSTRACT

Background:

Medical school curricula are managed through fragmented, static tools that prevent faculty and curriculum committees from accessing a unified view of content scope, sequence, and licensing examination alignment. Unintended content redundancy accumulates when course directors cannot see what colleagues teach, and curriculum governance relies on labor-intensive manual reporting that is neither timely nor consistent. No widely adopted platform incorporates a generative AI agent capable of querying a relational curriculum database and producing tailored, interpretable, and on-demand reports.

Objective:

This study aimed to design, implement, and evaluate a four-phase, AI-integrated relational curriculum mapping framework providing faculty self-directed, real-time curriculum search; cross-course alignment reports classifying content as redundant, complementary, or a gap with actionable recommendations; and institutional alignment with USMLE content and accreditation requirements.

Methods:

We developed and implemented a four-phase, AI-integrated relational curriculum mapping framework during AY 2025-2026. The framework captures session learning objectives (SLOs) annually, maps them to program objectives, course objectives, USMLE codes, AAMC keywords, and Bloom’s levels using a large language model with expert verification, centralizes verified data in a relational database (Airtable) with faculty read-only search, and deploys a second AI agent via API to generate on-demand reports covering scope, sequence, redundancy, gaps, and USMLE alignment across all 16 preclerkship courses and all clinical clerkship rotations.

Results:

Course directors gained self-directed curriculum access via Airtable’s Extensions search tool. Cross-course vertical alignment reports comparing multi-course sequences identified redundant, complementary, and gap findings and generated actionable, course-level recommendations, including session consolidation, cross-referencing conventions, and content-ownership assignments. A cross-course vertical alignment analysis of two preclerkship courses identified 50 matched SLO pairs across three USMLE pathology domains within minutes. A USMLE Step 1 alignment report across all 16 courses identified a curriculum-wide pathology coverage deficit (42% versus a target of 45%-55%) that was descriptively consistent with institutional examination performance data, though this association was not formally tested. An accreditation-focused query identified all preclerkship and clerkship content related to a specified topic and generated Data Collection Instrument-ready documentation within seconds. A full MEPO coverage analysis across all 9 clinical clerkships, generated within minutes, found 24 of 25 (96%) institutional program objectives had documented clerkship coverage and identified 1 objective with no matching SLO in any clerkship. The relational architecture supports LCME Element 8.3 requirements.

Conclusions:

An AI-integrated relational curriculum database and faculty-facing search interface can convert curriculum governance from a periodic, retrospective activity into a continuous, self-service one, producing alignment and gap analyses with specific recommendations within minutes rather than weeks. This can reduce staff workload. Extension of this tagging and reporting architecture to assessment items, and prospective correlation of alignment evaluation and assessment metrics with licensing examination outcomes, are planned as the focus of subsequent work.