Jon Gorecki
Makes things · breaks things · fixes things (mostly)
I make things. Music mostly, for a long time now, and more recently software. I play a lot of Magic. I'm in Chicago with my family and my dogs, and I'm happiest when something is in pieces on the desk and I'm working out how it goes back together.
None of that is my job. My job is being the person you call when something technical is broken and nobody can work out why, which I'm good at and genuinely enjoy, but it isn't the part of me I'd lead with at a party. That section is further down if it's what you came for.
Music
Birthdayy Partyy
I make dance music with my twin brother Bryan as Birthdayy Partyy. We started in 2018 and haven't really stopped: bassy tech house, drum and bass, and whatever else we talk each other into. We've played EDC Las Vegas, Electric Forest and North Coast, and we've put out tracks with people like Odd Mob and Adventure Club.
I've been an audio nerd since I was a teenager running live sound, which is where I learned that most of the job is listening carefully to something nobody else is paying attention to. That has turned out to be transferable to an embarrassing degree.
2018 the year Birthdayy Partyy started, and we haven't really stopped since
Magic
A game I can't leave alone
I've played Magic: The Gathering for years. It's a thirty-year-old game with a 300-page rulebook, thousands of cards that each rewrite the rules slightly, and a community that enjoys arguing about the edges. It is, in other words, exactly the sort of system I can't leave alone.
Two of the things I've built came straight out of playing it.
A machine that sorts my cards
Cardomancer · started November 2024I own an unreasonable number of cards and sorting them by hand is miserable, so I built a machine that does it instead. The physical build and the software both. A camera looks at a card, the software works out which card it is, and the machine drops it in the right bin.
Identifying a card sounds like a job for AI and mostly isn't. OpenCV finds the card in the frame and straightens the perspective so the image is square. The identification itself is image hashing, comparing that corrected image against a reference set. Vector embeddings give a second, independent opinion. When the two disagree, or either one is unsure, it stops and asks me instead of guessing.
Worth being precise about, because "AI project" gets used loosely: no language model ever looks at a card. The recognition is hashing and embeddings. What the AI did was write the code, with me directing it.
~99.7% accurate, across several thousand cards sorted
A rules bot, and the night it proved me wrong
Rulemancer · 2026Rules arguments at a table get settled by whoever sounds most confident. That person is regularly wrong, and it was regularly me. So I built something that searches the actual rulebook, answers the question, shows the rule it used, and says plainly when it can't answer rather than guessing.
The bot isn't the interesting part. The interesting part is what happened when I started measuring it.
I'd tested whether the rulebook text was doing any work, by swapping in rules belonging to a completely different question, and concluded it barely mattered. Worth about three points. Wrote it up, moved on. Then it occurred to me that nearly every question in my test set names a specific card, and a card's own text usually restates the rule it depends on. So I built a set of questions with no card names in them at all and ran the identical test.
My original number wasn't wrong, it just answered a question I hadn't realised I was asking. The rules aren't redundant, they're redundant when you already have the card text, and 99.4% of my questions had card text. The experiment that overturned it cost about five dollars, which felt like a bargain for not staying wrong in public.
That's the part I'd actually want to talk about. Building the thing is easy now. Working out whether it does what you think it does is the hard bit, and almost nobody bothers. It scores roughly 86% across all 1,409 questions, I checked the AI doing the grading in both directions, and the results page regenerates its own figures from the underlying data and refuses to publish if a number drifts from the run behind it. Because I no longer trust myself to notice.
98.84% → 15.12% what scrambling the rules costs once no card text is there to cover for them. Not three points. Eighty-three.
Work
The part that pays for the rest
Eight years at MultiTaction, which makes big interactive displays and the software that runs on them. I started on support and ended up as the senior person customers got handed to once their problem had already defeated somebody else. The title was Senior Customer Success Engineer.
Most of it was what it sounds like: something isn't working, someone is frustrated, and it's usually not the thing they think it is. Somewhere north of fourteen thousand of those. The ones I remember are the big installs, a sixty-site rollout for a bank and a thirty-display video wall for a medical center, where the stakes were high enough that staying calm was most of what I added.
The other half was teaching, and that's the half I liked. Onboarding, training, writing the guides people actually reach for, aimed squarely at making myself unnecessary to them. My read is that cut the repetitive back-and-forth on common problems roughly in half, though I'm going off watching case volume drop rather than anything I tracked properly.
When something was tedious enough, I automated it. Logging one hardware repair used to mean typing the same details into three places and hoping you'd attached the right serial number to each. People hated it, so people rushed it, so the data was wrong, so the reports built on it were wrong. I rebuilt it as a single action that creates the shipment ticket, the repair case and the replacement case, and links the correct assets to all three. It cut duplicate entry by at least 4x and errors to basically zero. It also took a few hours, which is sort of the point. Nobody had spent the few hours.
14,000+ support cases, give or take, over eight years
For clarity, since these titles blur: I was the technical point of contact and the escalation path, not the account owner. I didn't carry a number, own renewals or run QBRs. I'm aiming at technical account management and customer success engineering as a next step, not describing a job I've already had. I'm looking now, remote in the US or hybrid around Chicago.
Say hello
Get in touch
Happy to talk about any of this. Music, Magic, dogs, or whatever you're currently trying to get working. LinkedIn is the best way to reach me.