Twelve months ago I could not use git. Today there are forty public repos under my name with twelve thousand clones between them. This is my birthday month and the anniversary of my first real commit, so I am writing down how that happened.
I make pizzas at Jet's in Hamtramck for three hundred dollars a week. That is the budget. That is the entire R&D department. Everything you see in my portfolio got built around closing shifts.
I left the path on purpose
I studied computer science and I dropped out. Not because I failed and not only because of COVID. I watched what waited on the other side of the degree and I did not want it. My friends graduated into big companies and called me from the other side, miserable in good jobs, flat voices on the phone. The salaries fed bureaucracies and supply chains I do not believe in. I decided the money was not worth that trade.
So I went to Detroit and made music. Ran a record label, threw a lot of events. It taught me how to ship something real to a room full of people and live with the result. That is the same nerve you use to publish open source.
The machine that flipped me
Then I watched an LLM write code. Something landed immediately: this was a real shift, the kind that happens twice in a lifetime, and it was happening now. I did not want to read about it. I wanted my hands in it.
Here is the irony. I am an artist and I mostly hate AI. AI music and AI images turn my stomach. The old DeepDream stuff had something to it, and maybe a tenth of a percent of people use these tools to push art somewhere genuinely new, but the rest is slop. Code is different. It is tangible and functional, and I still get to run my art brain through the design and the out of the box thinking. So I can hate the machine and love building with it at the same time.
The deeper reason is a bet. I see a future where local models are as capable as the bloated ecosystem destroying ones we run today, and when that day lands I want to be in the driver seat, not renting one. I am careful with my money. OpenAI and Google will never see a cent of mine, and I hate that Anthropic's models get pointed at the US military. But $20 a month is a drop in the bucket against the future I see coming, and it is the only software I pay for. I have always hated SaaS and rented tools. The way out, the part I actually believe in, is building open source and giving the knowledge away.
I started with tiny Python scripts, things that drew on screen in twenty lines. Then I tried to build a game in pygame before I understood any of it. The game grew. The game broke. It kept breaking.
That broken game is the most important thing I built all year, and you will never play it.
It is where I learned git, because I needed to stop losing work and start walking backward when I wrecked something. It is where the abstraction stopped being magic and became a system I could reason about. I got bit. I stopped making music. I started building every day.
Months later I came back to it. The bugs I could never get to the bottom of the first time, I finally killed, and I compiled the whole thing into one exe you could double click. That moment seeded something. It is most of why I love Go now. Write it, build it, ship a single file, nothing to chase at runtime.
I learn by breaking things
I have always been like this. Lectures and videos do not stick. I learn by doing the thing, getting it wrong, and going again. AI is the perfect tool for that. It let me reach three levels above my paygrade, watch it fall apart, and find out why. From there the depth compounded fast. Frontends, backends, APIs, databases, vector search, local model inference, Flutter and Kotlin on mobile, a transpiler, Rust and Zig and Go. In a year I have worked across the whole stack except the silicon. No FPGAs, no deep hardware. Everything above that I have been inside.
I trained my own model
I trained my own small model. Not from scratch. I took a Qwen 3.5 0.8B base and ran it through two passes. First I fed it reasoning traces pulled from every SOTA model I could get, paired with coding sets. Then I benchmarked what it could actually do and trained it again only where it lagged. Most people distill one big model down into one small one. My bet was that mixing all the frontier models would give better generalized reasoning, and it worked. The result outranks the base in every category I tested. It is not releasable, more a proof of concept, but it writes the summaries in my research digest and powers my code blog. No API key, no token bill, no company in the loop.
Efficient, on twenty dollars a month
Call it vibe coding. I will take the label. My friend Tommy, who is plotting a company called a5ecreta9ency, five nines of uptime, says when it exists he is making me Lead Vibe Coder. I will take that too, with one correction: I am efficient at it. Everything here was built on free tools and a single $20 a month plan. No $200 subscriptions, no funding, no team, no cloud invoice. Forty repos and a trained model out of pocket change. Constraint is the method, not the excuse.
Local first, because I always have been
This part is old. I ran Minecraft servers in middle school. I have always preferred owning the box to renting it. So when I learned Docker, then Proxmox, then how to run a real homelab on bare metal with a Debian container per service, it was not a pivot. It was permission. The kid who wanted his own server finally had the skill to self host an entire stack and a reason to: software ownership should not cost a subscription.
That is the whole philosophy. Offline first. Local first. No gatekeepers, no paywalls, no cloud when it can be avoided. Tools that keep working when the internet does not. Every repo is the same argument made another way. A local AI gateway so I get my own vLLM without losing a night fighting my junk hardware to make it run. A social feed that replaces Instagram and does not run on your anger. A case study where I took a real client off a $132k a year AWS bill, done free, no NDA, just for the reps. They ghosted me after they took the plan they loved. Their site moved anyway, so I think it shipped. When you cannot afford the cloud, you build things that do not need it.