● Ryan Carter · Creative Technologist, Toronto

Selected Work.

CH.01 · Broadcast2005 – 2023

Bell Media · MuchMusic · MTV · Crave · CTV · CTV Comedy Channel · TSN · Virgin Radio · CP24

Award-winning Creative Writer · Producer · Editor · Director.

Miley Cyrus at the MMVAs

Creative Writer · Producer · Editor
MuchMusic

eTalk at The Oscars

Creative Writer · Producer · Editor
CTV

Marilyn Denis · Final Farewell

Creative Writer · Producer · Editor
CTV

Party Smart PSA

Creative Writer · Producer · Director · Editor
MuchMusic

CTV Throwback

Creative Writer · Producer · Editor
CTV

Hockey Canada PSA

Creative Writer · Producer · Director · Editor
TSN

We Day Promo

Creative Writer · Producer · Editor
MTV / MuchMusic
CH.01b · Archive
  1. 01 CTV Comedy Channel Movie Promo CTV Creative Writer · Producer · Editor Watch ↗
  2. 02 CTV Comedy Channel Launch Promo CTV Creative Writer · Producer · Editor Watch ↗
  3. 03 MM Top 20.ca MuchMusic Creative Writer · Producer · Director · Shot · Editor Watch ↗
  4. 04 Dear Rouge · That Much Closer MuchMusic Creative Writer · Producer · Director · Editor Watch ↗
  5. 05 MUCH Freeview MuchMusic Creative Writer · Producer · Editor Watch ↗
  6. 06 MUCH Holiday Wrap MuchMusic Compositing · Visual Effects Watch ↗
  7. 07 Your Morning Promo CTV Creative Writer · Producer · Editor Watch ↗
  8. 08 Virgin Radio Image Spot Virgin Radio Creative Writer · Producer · Editor Watch ↗
  9. 09 Seinfeld · 10-Second Cut CTV Comedy Creative Writer · Producer · Editor Watch ↗
  10. 10 CTV Comedy Channel Movie Promo (earlier) CTV Comedy Creative Writer · Producer · Editor Watch ↗
CH.02 · SJC Central Video2025 – present

Chatelaine · Maclean's · Toronto Life · FASHION · Today's Parent · Canadian Business

In November 2025 I initiated and now lead the Central Video team at St. Joseph Communications.

Three shooter/editors report to me, and I shoot and cut alongside them. I designed and own the technical workflow, project management and media management that let the group operate as one team.

Within the first six months the team had directly enabled or supported over $1M in revenue.

The Group Chat

Directed · Edited
Chatelaine · Ep. 8 · Meredith Shaw

Canadian Business Innovation Awards

Sr. Videographer · Creative Director
Canadian Business · event sizzle
Details

The screens running live in the room — a three-screen wall behind the main stage. The same artwork cut for that wall, for elevator screens, and for broadcast sits in CH.03.

Maclean's × The Wealthy Barber

Director · Team lead · Editor
Maclean's · long-form
Details

The long-form interview the short cuts are pulled from. The transcript-to-rough-cut process behind those cuts is Case 03 in CH.03.

CH.03 · AI production2024 – current

"I want to use the tools we didn't have yesterday to solve the problems we couldn't solve yesterday."

me

Paint-out and cut selection.

Group Chat frame after the mic-stand logos were painted out The same frame before, with mic-stand logos visible Painted Original
Where AI helped
Painted out curtains and mic-stand logos. Picked social clip candidates from the transcript.
Where AI did not
Editorial voice. Cut rhythm. Guest curation.
Tools
Photoshop generative fill.

One element, every ratio.

Broadcast promo · 1920 × 1080 · 16:9
Stage wall · 3840 × 1080 · 32:9
Elevator screen · 2160 × 3840 · 9:16
Where AI helped
Motion and sparkle on background elements. Regenerated a single base across ratios and formats — pre, during, post.
Where AI did not
Design language. Palette. Live orchestration.
Tools
Runway ML · Adobe Premiere.

Transcript → rough cut.

Maclean's cut · The Wealthy Barber Maclean's cut · satellites Maclean's cut · building fast and cheap Maclean's cut · Alberta and Ottawa Maclean's cut · AI in the cubicle Maclean's cut · Mark Carney at Davos
Six Maclean's cuts from the same process, including one off the Wealthy Barber interview above. Each opens on TikTok.
Where AI helped
A-line assembly from Zoom (15–20 min source → 1–2 min social cut). Clip picks from long-form interviews (Maclean's × Wealthy Barber, half hour → 16:9 cuts).
Where AI did not
Dialogue smoothing. B-roll. Timing. Tone.
Tools
A custom Gemini gem · Claude.

All-generative, on purpose.

Node graph of the generative pipeline: character reference, shot prompts, video generations, stitch
The pipeline · character lock → shot prompts → generation → stitch
Where AI helped
All visuals. Character, animation, style consistency, backgrounds, transitions.
Where AI did not
Script. Voice. Pedagogical arc. The decision to be illustrated.
Tools
Runway ML (node-based).
Case 05 Client work · EN → FR

English to French, at roughly 1/15th the production cost.

KitchenAid spot delivered in French, with translated end card and supers
Where AI helped
Transcription, translation, and SRT alignment. Catching mis-transcriptions across languages that would otherwise ship.
Where AI did not
Final voice direction. Native-speaker cultural review. Client sign-off.
Result
A $30,000 content job delivered for roughly $2,000 internally.
Tools
Adobe Premiere · Claude.
CH.04 · Augmented RealityPhone-camera AR

CP24 · MuchMusic · CTV

Spatially aligned media, on almost any real-world surface.

Point a phone at something physical — a transit shelter, in-store signage, print, an endcap — and tracked media plays on it. The surface never changes. What someone standing in front of it sees does.

MuchMusic lens: a title card reading What VJ Are You? pinned above the wearer's head

What VJ Are You?

Built
MuchMusic · head-tracked lens
Details

Tap and it spins through the VJ roster, landing on one. The card stays pinned to the wearer's head as they move.

Instagram, built in Spark AR — Meta's platform, since retired.

Amazing Race Canada lens: a 3D pack and CTV badge registered to the wearer's body
The same lens holding registration as the wearer raises an arm

The Amazing Race Canada

Built
CTV · body-tracked lens
Details

The pack and badge are registered to the body, so they hold as the wearer moves — the two frames are the same lens a few seconds apart.

Instagram, built in Spark AR — Meta's platform, since retired.

CH.05 · Internal toolsBuilt for SJC

Two tools that removed a subscription and a bottleneck.

Tool 01

Caption Studio

Gives the video team an owned caption workflow instead of another SaaS dependency. Transcription runs locally through Whisper — no upload bottleneck and no per-minute API bill.

Built with
Claude
Runs on
Python · Whisper (local) · ffmpeg
Used by
The video team
Tool 02

Magazine Studio

Lets anyone on any team effortlessly turn a PDF into a 3D magazine-style flip-through. Drop the file in, get a page-curling issue back.

Built with
Claude
Runs on
Python · ffmpeg
Used by
Anyone — mostly art directors who create the PDFs
CH.06 · Innermap · CTO2023 – present

It doesn't get much higher stakes than a person's wellbeing. It doesn't get more controversial than someone genuinely connecting with AI.

  • CTO
  • Product · Engineering · Evals

Innermap builds Affi, an AI companion grounded in Emotion-Focused Therapy. The premise is that most people aren't short of coping strategies — they're short of a clear view of the patterns underneath. Affi works on what's underneath, and maps it as it goes.

Doing AI right. Doing AI safely. That's what Innermap is built around, and what I've been responsible for turning into product and engineering solutions as CTO.

I lead product, engineering and evals. Clinical direction sits with registered psychologists. Nothing user-facing ships without clearing the evaluation pipeline.

Everything user-facing passes through the evaluation pipeline before it ships. See CH.07.

Meet Affi

Product · Engineering · Evals
CH.07 · Governance · Ethics · Security

When AI arrived I asked what the most valuable skill would be, now that so much of the work could be done faster than we could do it ourselves. I didn't have an answer then. Three years on, I'm confident I do: testing.

Automated tests can cover the code. But what covers the murkier case — when the response coming back may one day need to talk someone back from a very dangerous edge?

Failure breakdown for one conversation state, with its score distribution
Failure located by conversation state, not just totalled

That is why I built the response validation system early. A separate LLM grades every response against a 214-metric rubric supplied by our clinical team. And because that grader will not be right at first either, human feedback is folded back in — a thumbs up or thumbs down on individual judgements.

Active criteria version v1.1, checksummed, 214 criteria split 130 harm-critical and 201 therapeutic
The rubric is a versioned, checksummed artifact

It is far from perfect today. It is built so that with use and iteration it measurably improves, release over release.

Flow version comparison table showing evaluations, harm issues, harm rate, cost and duration
Harm rate across flow versions · 23.1% → 14.0%

Validation runs on my local system. It deliberately does not run in production — that would mean putting protected health information in front of an evaluator, and it isn't worth it.

Human feedback calibration: 11 total, 45.5% accuracy, five accurate against six too harsh
The judge is judged · 6 of 11 human checks called it too harsh
Innermap · security posture
  1. 01
    Encryption at restFernet
  2. 02
    Encryption in transitAll user-facing traffic
  3. 03
    AuthenticationAuth0
  4. 04
    HIPAA-oriented infrastructureTechnical safeguards implemented · BAA pending
  5. 05
    Evaluation gatesAutomated + human sampling, pre-release
Validator · Affi

Every generated message is scored against a versioned rubric. Harm-critical is a flag raised by that rubric, not a reported incident. These runs are pre-release evaluation on my local system, not observed production traffic.

Validator dashboard: active criteria version v1.1 with checksum, 214 criteria, and evaluation totals
The rubric is versioned and checksummed · v1.1 · 214 criteria
Failure breakdown by conversation state with per-state score distributions
Failure located by conversation state, not just totalled
Flow version comparison table showing evaluations, harm issues, harm rate, cost and average duration per version
Regression across flow versions · 23.1% → 14.0%
Human feedback calibration comparing thumbs up and down against the model's own evaluation verdicts
The judge is judged · 6 of 11 human checks called it too harsh
Cost and performance metrics: total cost, evaluations, average cost per evaluation and duration
$0.0004 an evaluation · cheap enough to run on everything

There's never been a better time to be a Creative Technologist.

The tools will surely keep changing. The mediums seem to multiply every day.

Creativity will leverage AI. AI will enable innovation.

But the spark of authentic human imagination will never be replaced.