The demand-side companion to the tool directory: what the Meridian community has asked to do with AI, how each request maps to a context profile, and what was decided.
Instructors may use AI to draft formative feedback on submissions, provided the instructor reviews and owns every comment and the syllabus discloses the practice.
Students may generate summaries, flashcards, and practice questions from their own notes and published class materials — subject to each course's AI policy.
Fully automated grading was declined: the Teaching & Learning profile requires a human grader of record. Requesters were routed to the approved feedback-drafting use case instead.
Began as a registry request from IT Operations; approved, then graduated into a configured deployment with its own service and system cards.
Permitted for routine, template-style correspondence only. Out of scope for employee-relations, disciplinary, or medical matters; inputs capped at Internal data.
Proposed for committee and project meetings. Open questions on participant consent and records retention are with the governance committee.
A high-risk request touching applicant PII: approved with added controls, then graduated into a deeply configured CRM deployment with risk evidence behind it.
A proposed self-service chatbot for common advising questions. Escalation paths to human advisers and accuracy thresholds are still being defined.
Research teams may use AI to screen and rank published literature, with screening criteria and prompts documented in the study protocol.
Permitted only on the Secure Research AI Cluster, with an approved IRB protocol and data-use agreement per project. Never on general-purpose services.
Every entry follows the same path, so a decision made once doesn't have to be re-litigated tool by tool.
Anyone on campus can propose a use case through the request process. The registry records it even before a decision exists.
Each request is assessed against the relevant context profile — teaching and learning, research, administration, or student engagement.
Approved, conditional, or declined — the decision, its conditions, and its date stay on the record. Declined entries remain visible so the reasoning isn't lost.
A use case that hardens into a configured deployment gets its own service or system card in the tool directory, with risk evidence behind it.