Why a dancer builds AI pipelines

The ending

In 2015, I was a professional dancer with the National Ballet of Canada — a 12-year career, national and international productions, the thing I'd worked toward since I was a kid. Then I herniated two discs in my lower spine: L4-L5 and L5-S1, with a fragment pressing on a nerve that made my left leg unreliable. The pain in the first month was excruciating.

I took six months off to heal and try to return. I managed a brief comeback before L5-S1 opened back up and the herniation came back. I offered my resignation, but the company did something I didn't expect: they fired me on the spot and gave me a severance larger than I would have negotiated myself. It was the best financial outcome, even though it read backwards.

What followed was years — long, very lonely years — of depression, isolation, and addiction. I locked myself in a basement. A 12-year relationship ended. I sold off property I owned. I cashed out retirement savings to qualify for social assistance because I couldn't find work and I didn't believe in myself enough to try. I couldn't afford school to retrain, and nothing that was available even appealed to me. I was stuck in a way that felt permanent.

The only option

At some point, things couldn't get worse. I won't detail what that looked like, but suffice it to say: a change was necessary. And without any real consideration, I decided the only thing I could do was go to school. Do something, anything, with my time that pointed me in a direction other than the one I was facing.

The only program I could access by transit that appealed to me was jewellery and metalsmithing at George Brown College. It aligned with my need for physical work (I'd always needed my hands to move), it served as a creative outlet, and it connected to some wire-sculpture work I'd been doing at the time. I figured: whether I got work doing it or not, at least I might enjoy it and get better at something.

I did enjoy it. But my time there was marred by a collective bullying dynamic — from some students and a few faculty members. At first I thought I was being paranoid. After six months and multiple signs, I understood what was happening. I left mid-second semester, with no blame assigned, no hair raised. It was just time to go.

The real hobby

Long before George Brown, I'd been obsessed with building computers. Not as a job — as a way to stay sane.

When my mind would race and I couldn't get out of my head, I'd build something. Crochet did this. Wire sculpture did this. And then, computers. I was obsessed with subverting Apple's imposed hardware restrictions — the fact that you pay 300% more for components just because they're in a white plastic case with a logo on it. The moment I discovered you could install macOS on bare-metal generic PC hardware with an Intel CPU, I was all in.

I built three Hackintoshes for myself. I built one for my ex when he needed something more powerful than a laptop but couldn't afford a $3k iMac — I built him one for under $500 that beat the machine he was eyeballing in benchmarks by a wide margin. I turned my Dell Latitude into a MacBook. I sold a couple to people who wanted Mac-like systems but were also broke. I got so efficient at assembly that I once had a guest come visit, we got high, I started building a computer, and I had the machine built, powered on, and outputting video before the drug even started to hit — usually 30 minutes in.

The reason I got into this was simple: I wanted to know why these things were so expensive to buy from a store. So I bought the components and did it myself for cheap. I learned to fix things because I had to debug hardware conflicts. Software too. The result: this stuff is massively overpriced and people pay way too much for it by relying on someone else to do something they can do easily. Even easier now with language models to help.

The bridge

I needed money. I needed a hobby. I figured I had a decent enough understanding of English, a critical eye to spot problems in images, and enough confidence to back up my positions with logic instead of vibes. So I started doing AI data annotation — evaluating AI-generated images, labelling bounding boxes, assessing LLM outputs against rubrics. I didn't do it to understand AI better, but I certainly began to.

I stopped looking at these models like perfect systems with all the answers. I started understanding them as highly accurate prediction systems that rely on an amalgam of very imperfect human data. When you see how they fail during reinforcement training, you learn a lot about how they actually work.

In conversations with a wide variety of models about career suitability — based on skills I was developing, hobbies I was dabbling in, things I was interested in — they almost unanimously pointed toward project management, studio assistant work, creative instruction, or creative technology. Given my artistic background and professional history, paired with me experimenting with ComfyUI and testing Google's IT cert courses, I figured something hybrid like creative technologist might be the right direction.

The problem I saw

But there was a glaringly obvious issue: I have no professional experience in this field. No formal education. And I cannot write code from scratch.

Yet somehow I've managed to build several utilities that work — and I think faster than someone who can code would have built them on their own — by getting an assist from an LLM. It started with a Python script to clean up dataset images in bulk. Then it shifted to a tool that pulled the most-used scripts together in one place. Then a UI so anyone afraid of Python could use it. Then I thought: why rely on Python and Inkscape when I could dig out the components, rewrite them in Rust as a cleanroom implementation, and replace all that dead weight?

And it was mine. I was responsible for it. I had to try it, find bugs, relay the issues to an LLM, decide on actions to take, then make it happen and try again. I discovered that having models with different genetics, different priorities, different ideas of "optimal" — models with different blind spots and hyperfocused strengths — made a massive difference in how long I'd have to spend debugging. With the same model, it would take nearly forever and I'd end up with broken, janky code. With differing perspectives, I could throw together working code in no time.

But I'd never documented this learning process. I'd never written down the steps I took and why. So I decided I needed to do something relevant to a creative-tech role: start from scratch, walk through the creation, and document what was happening and why.

The project that changed things

I also wanted to create a storybook — an illustrated poem for adults. And doing it manually with existing free tools was sticky. So why not create a more lubricated system? One that's adjustable but not overwhelmingly so. That's where MiniBook came from.

It's a dual-LLM pipeline I designed end-to-end. I specified the logic, directed LLMs to implement it, debugged the output, and iterated until it shipped. One prompt in. A finished, illustrated, character-consistent PDF out. The engineering judgment — knowing what to build and why — is mine. The implementation was AI-assisted. That's how I work.

And it's not a one-off. It's proof of a skillset that's rare right now: the ability to direct AI to build working systems and exercise genuine judgment over the result.

What I'm looking for

Being the most useful person I can be is my biggest goal. I've always gravitated toward roles that let me use my talents and unique thinking to bring a concept or project to life — something greater than what I could ever conceive on my own.

That's why I loved being a dancer. Working with a choreographer was a dream come true. They asked for something. I tried to do it. We iterated, tweaked, perfected, and then presented. What I contributed to that choreography was so much more than what I could have conceived on my own — but my creativity and problem-solving style was a measurable part of the bigger picture. Much bigger than if I were the choreographer alone or had to do it myself.

That kind of working relationship. That environment. That satisfaction about contributing to work I believe in — that's what I consider success. What that looks like in a job title, I don't know yet. But I know what it feels like, and I know how to recognize it when I see it.