Content Designer · Vancouver, BC · open to remote (Canada)

I make complicated products make sense.

Instant clarity. Honest language. Systems that scale.
Content design for the products with fine print — AI, licensing, security.

See the work
Illustration: a tangle of interface scribbles on the left; a person walks right, drawing one clean teal line that arrives at a tidy, organized interface.
GETTY IMAGESSAMSUNG KNOXAUTODESKSHOPIFYLEMONADE 2018 · contractBEST BUY

Words for every kind of complicated.

Pick a problem. There’s a case study for it. Interfaces below are recreations — the words are what actually shipped.

Skip ahead — read the full portfolio →

Recreated product dialog: Use this image as a reference? — with the CTA pair Generate now, license later and License now.
Getty Images · iStock

Making AI image generation feel fair, not sneaky

Getty’s AI generator lets you start creating with photos you haven’t licensed yet — on a site that sells licenses. I wrote the copy that keeps the eventual bill honest, visible, and never a trap.

86% comprehension in testing · shipped on two brands
Read the study →
Recreated download options: Higher resolution and Enhanced detail, each with a What to expect expander written at equal length.
Getty Images · iStock

Telling users the AI might change their picture

The most popular download option ran AI enhancement that could subtly alter an image — smooth a texture, shift a face. I rewrote the download copy from selling to describing, both options at equal length.

Copy delivered verbatim to engineering · 1 production bug caught
Read the study →
Recreated assistant message: Your message could not be processed because it may violate our content policy. Please rephrase your request and try again.
Samsung Knox

Writing the “no” for a security AI

When an AI assistant inside a security console refuses, fails, or asks permission, the wording decides whether admins keep trusting it. I wrote all thirteen of those messages.

13 written · 9 shipped in beta
Read the study →
Terminology audit excerpt: EMM, UEM and MDM logged as the same concept across different consoles, with match-type tags.
Samsung Knox

Getting seven products to speak one language

Seven admin consoles had drifted into seven dialects — three different names for the same thing. I catalogued ~6,000 terms and built the glossary to unify them, including an AI-assisted pipeline that failed twice before it worked.

~6,000 terms · 184-entry glossary in review
Read the study →
Evaluation rubric scorecard: five dimensions scored 0–3, totalling 9 of 15, root cause state management.
Autodesk

Measuring whether the AI is actually good

Everyone knew the AI support assistant was “sometimes wrong.” Nobody could say how, how often, or in which ways. I built the evaluation framework that answers exactly that — and caught the judge AI hallucinating mid-audit.

15-point rubric · piloted on 702 real conversations
Read the study →
Archival experience map for Bangor University prospective students, in an ink and teal duotone treatment.
Domain7 · 2019–2022

The agency years, preserved

Experience maps, personas, and content strategies for Carbon Trust, Bangor University, and Concert Properties — kept because they still argue well.

5 artifacts on the shelf · 61% → 89% navigation success
Browse the archive →

There’s also a strategy & leadership shelf — setting direction, building practices, democratizing experimentation. Four more studies →

The receipts.

~6,000
UI terms unified under one governed vocabulary — a framework other designers now run
7
AI products shaped since 2018 — from an insurance chatbot to agentic enterprise assistants
12 yrs
of making complicated things make sense
3
enterprise AI teams — trust, language systems, and evaluation — through my independent practice

Every number traceable to a project above.

I’ve been teaching AI to talk since 2018.

It started with scripting a chatbot’s lines, six years before the wave. These days it’s response systems and runtime contracts. Same lesson — smarter student.

Illustration: a person on a step-stool edits a robot's giant speech bubble — a crossed-out scribble replaced by one clean teal line.
An AI, probablyHi! I’m sorry, your request contains content that violates our policies.
ZameerYou just accused someone of a crime. They typed “reset my password.”
An AI, probablyOkay. I panicked. What should I have said?
Zameer“Your message could not be processed because it may violate our content policy. Please rephrase and try again.” One word — may — does all the ethical work.
An AI, probablyGot it! Your changes are live! 🎉
ZameerWhat did the confetti earn? You’ll say this fifty times today. Nobody survives fifty parties.
An AI, probablyEdit applied.
ZameerBeautiful. Two words, noun + past participle. You’re learning.

How I work.

1. Audit the chaos

Every term, every screen, every claim — captured exactly as shipped, before fixing anything.

2. Build the system

Glossaries, taxonomies, response classes, frameworks — the structure that decides which words can exist.

3. Ship the honest version

Copy that tells the truth about what the product does — especially when the truth is complicated.

4. Leave the system behind

Glossaries, frameworks, runtime contracts — built to keep working after my engagement ends, and socialized up to the executives who fund them.

Illustration: a person hands over a single small card while a tall stack of bound, bookmarked volumes stands behind them.

This is the short version.

The full portfolio goes deep: eleven case studies with interactive exhibits, an archive of the agency years, and a ledger of everything since 2010.

Read the full portfolio →

Roughly a 12-minute read. The good details live there.