Outline Engine: An AI-Powered Instructional Design Tool
Instructional designers lose hours turning policy docs and SME transcripts into something teachable before any real design starts. Outline Engine does that first pass: objectives, a storyboard, script text, quiz items. Every claim cites the source it came from, and every field is editable and can be commented on, because the point is to hand you a draft to argue with.
Audience: Instructional designers and L&D teams
Responsibilities: Instructional design, AI prompt engineering, dashboard UX design, information architecture, front-end development, brand identity design
Tools Used: Claude, Claude Design, API Logic, Google Docs
The Problem - Hours lost before design starts
Every course I’ve built starts in a similar way: a policy document, a messy SME transcript, or a product spec with a deadline attached. The material is real. It just isn’t shaped for learning yet. On the last compliance module I wrote, I spent most of two days reading a 40-page bank security policy before I could write a single objective. That translation step eats time I’d rather spend on pacing and on whether the practice actually builds the skill.
Working with AI - Designing for doubt
Early versions gave me objectives that were just source headings with a verb bolted on. “Understand the expense reporting policy.” Useless, since you can’t watch someone understand. I had to spell out that an objective describes something observable, and hand it Bloom’s verbs to work from.
Quiz distractors had the reverse problem: the wrong answers were so obviously wrong that no one would pick them. Two changes fixed it. I asked for distractors pulled from elsewhere in the source, real rules applied to the wrong situation, because those are the ones people actually fall for. And I banned “always” and “never,” which helped more than I expected. Learners eliminate absolutes on reflex without reading the rest of the option.
The bigger realization came from asking the model where things came from. When I required every objective and quiz item to cite the chunk of source material behind it, I could finally see what it had invented. My source folder had two versions of the same policy, one of them superseded, and it cited the old one. Without the citation I’d have shipped the outdated threshold. That constraint shaped the prompt too: asking for structured output rather than prose is what let each field carry its own source and stay independently editable.
Then I added comments. Reviewing a draft, I kept wanting to flag things I couldn’t resolve yet. Check this number with the SME. This policy changed last spring. Is this even the right audience. Those notes were living in a separate doc and getting lost, so every field takes an inline comment now. It isn’t an AI feature at all. It’s what makes the AI’s output usable, because a first draft you can annotate is a draft you’ll revise instead of rewrite.
Reflection - A first pass, not a replacement
I half-expected this to feel like cheating. It doesn’t. It’s good at the boring first pass and bad at knowing what a learner actually needs, which is roughly the split I’d want.
Two things I’d fix. It doesn’t know who it’s writing for: the same policy produces the same objectives whether the audience is new hires or people who’ve done the job for ten years, so the next version should take a learner profile before it takes source material. And alignment is still loose. The quiz items are decent on their own, but they don’t always measure the objective they’re sitting under, and the tool can’t see the gap. I’d want the objective and its assessment items generated together and shown side by side, so a mismatch is obvious at a glance instead of something I catch on a third read.