Case Study - Cutting Course Creation Time by 10x

Overview

A tutor on Edify teaching a new module used to start with a blank page, a stack of source material, and a calendar blocked out for the week. Multiply that by every new institution being onboarded — and by every assignment that then needs to be graded across classes of 60+ students — and content creation becomes the bottleneck for the entire platform. Off-the-shelf AI writing tools sit outside the LMS, can't see Edify's own library, and produce generic content that an instructor has to rewrite to match an institution's standards. On the grading side, a tutor running a class of 60+ students can lose eight hours a week to feedback alone, and that work is mostly mechanical pattern recognition that AI is uniquely good at — provided a human stays in the loop to set the bar and own the final mark.

The foundation is a knowledge index seeded with 100+ existing Edify courses and their high-performing assignments, tagged with metadata so retrieval consistently surfaces on-brand, on-standard reference material. On top of that we built four generation services and a review surface that ties them all together. A Course Generator turns a topic and learning objectives into full modules, lesson notes, and video recording scripts — generating a six-week course in under five minutes. An Assignment Creator produces prompts with rubrics scoped to multiple assignment types so an instructor can diversify how they assess students without spending the design time. A Quiz Generator writes Bloom's-aligned multiple-choice questions with distractors and answer explanations drawn from the instructor's own course materials. A Grading Assistant evaluates student submissions against a rubric, identifies strengths and improvement areas, drafts personalised feedback per student, and batch-processes up to 60 submissions at a time. Every AI output lands in a review interface where tutors can edit, regenerate, or approve — with full version history — before anything reaches a student.

Every generation call retrieves context from the indexed course library before the model writes a word, so the output sounds like Edify rather than like a generic AI tool. Content is tagged by subject, level, course rating, and NUC compliance status — the metadata that retrieval needs to consistently surface on-standard material. The human-in-the-loop pattern is non-negotiable: the AI proposes, the tutor disposes. Tutors retain full authority over what reaches students, and approved-vs-edited telemetry feeds back into improving the system's output quality. The same foundation positions Edify to package and resell pre-built certified courses to new institutions, turning the AI layer from an internal productivity tool into a revenue stream tied to the rate at which the content library grows.

What we did