I design the practice, not the presentation.

I'm an instructional designer working in scenario-based learning and AI-powered learning tools. Most training tells people things. The work I care about makes them decide something and live with it.

Storyline and Rise builds, branching scenarios, and custom software when the off-the-shelf tool won't do the job.

Professional Diploma in Digital Learning Design, Digital Learning Institute
MEd TESOL · sixteen years teaching

Storyline · Rise · Bubble · Anthropic API

Work

Built — case study

PrepPal — putting a quality gate on AI-generated learning content

PrepPal generates reading worksheets for tutors teaching academic English to university students. A good one takes a skilled person a couple of hours to write. Generated, it takes two minutes — but someone still has to read every question and confirm every answer, and that checking is the part nobody budgets for. The mistakes don't look like mistakes: they're reasonable-looking questions with the wrong answer attached.

PrepPal is a working system I designed and built end to end to test the obvious response: put an automatic check between the generator and the output. The Bubble interface and its workflows, the API integration, the prompt architecture, the document renderer and the validation layer.

It began as a managed build: I wrote the spec, ran the project and worked with a developer who got it off the ground — but the prompts turned out to be where the pedagogy lives, and writing them needed someone who knew how academic English is actually taught. That wasn't his field. I took the build on and did both jobs.

Produces a complete reading worksheet pitched at a set difficulty level — source text, comprehension items, a reading-skill focus, target academic vocabulary, seven graded tasks and an answer key — in about two minutes
Renders four deliverables per run: learner and facilitator editions, PDF and Word
Validates before it renders: vocabulary checked against the Academic Word List, and a second model pass independently re-marks every answer key

PrepPal configuration panel showing discipline, student level, skills focus and topic fields — The tutor sets discipline, level, skills focus and topic. Every value shapes what gets written, rather than filtering a template.

Two pages of a generated worksheet: the reading article with target words in bold, and the teacher page showing task timings and the answer key — The reading text with target vocabulary in bold, and the teacher page — suggested timings per task, notes on the open-ended activities, and the full key.

PrepPal generated material screen with tabs for student content, teacher notes and answer keys, and a box for describing a change — The teacher reads it back before committing to it — student content, teacher notes and answer keys on separate tabs — and can describe a change in plain English rather than editing by hand.

PrepPal saved materials library with download buttons for teacher and student editions in PDF and Word — Four files per run. The two editions share the same worksheet pages, so they can't drift apart — the teacher version carries the notes and key at the back.

Read a generated worksheet in full (PDF, teacher edition) — Business & Management, B1, produced in a single run.

Bubble · Anthropic API · Pexels API · Python / Flask · ReportLab · python-docx · Render

Design note Each reading worksheet carries twelve target words from the Academic Word List — a published list of the vocabulary that recurs across university writing — alongside the comprehension and reading-skill work. Words that weren't on the list kept coming through anyway — enrol, infringe, graph, mislead — across three unrelated subject areas. My first fix was a ban list in the prompt, which worked exactly once per word: a denylist only ever contains the mistakes you have already made. So I replaced it with a membership check built from the published list itself, 570 word families and 3,111 forms, running on every worksheet and blocking the render before anything could reach a student. Off-list targets went to zero, and have stayed there on subject areas the check was never tuned against. The lesson I'd take to someone else's project: when a rule can be checked against a source, check it there — not in the prompt.

In build — autumn 2026

The conversation you keep postponing — an AI-driven simulation

Self-directed brief. Client fictional; design problem real.

A company promotes a wave of first-time managers out of technical roles. One of them has a problem: their strongest engineer has started missing deadlines and deflects every time it's raised. The learner is that manager, and the conversation has to happen today.

The colleague's replies are generated live rather than scripted, so the learner responds to the conversation they actually get instead of hunting for the approved option. It closes with feedback written against what was really said.

The design problem I'm most interested in is the one underneath: a generated character will drift. Left alone it softens, forgives too quickly, and rewards a learner for being pleasant rather than effective. So the simulation runs a second check on every turn — is this person still behaving like this person?

Articulate Storyline 360 · Anthropic API · generated dialogue · character validation

In build — autumn 2026

Before you have the conversation — a manager's toolkit

Same fictional client. Companion piece to the scenario above.

Short, self-paced, and built to be opened twenty minutes before a real one-to-one rather than worked through once and forgotten. Covers separating observation from interpretation, opening lines that don't trigger defensiveness, and what to do when the answer is "I'm fine".

Deliberately not a course. The scenario is where people practise deciding; this is what they reach for afterwards, on a phone, in a corridor.

Articulate Rise 360 · self-paced · mobile-first · job aid

Background

I came to instructional design from sixteen years of teaching, most of it in higher education — which means I have watched an unusual amount of learning material meet a real room and fail. That's the useful part. I've marked the work, so I know which instructions get misread. I've run the session, so I know what fifteen minutes of preparation actually has to achieve.

Formally: a Professional Diploma in Digital Learning Design from the Digital Learning Institute, and an MEd in TESOL. Practically: I design for the moment someone has to do the thing, not the moment they're told about it.

I also write my own software when the tool doesn't exist, which turns out to be a useful thing for a learning designer to be able to do — and increasingly the thing people want to talk to me about.

Available for freelance learning design.

Storyline and Rise builds, branching scenario design, and help putting AI into a content pipeline without losing control of the quality. If you have a performance problem and no course yet, that's the conversation I like most.

sally.tallant@gmail.com