This project goes back further than I sometimes remember.
If I trace it all the way to the beginning, it probably starts around 2020 or 2021, shortly after I first got into affiliate sites. Around that time, I came across a piece of software called Puzzle Book Mastery. I don’t remember exactly where I saw it. It could have been Warrior Forum, JVZoo, or even an ad inside Keyword Atlas. Either way, when you’re poking around online projects, you constantly run into things like keyword tools, “make money” software, puzzle book builders, coloring book generators, all of it.
Puzzle Book Mastery was one of those.
If I remember right, it was something like $27 or $37. A base product with one or two upsells. I don’t think I bought everything, probably just the base and one upgrade. It came with several puzzle types, and one of them was a Sudoku generator. Easy, medium, hard. It produced printable PDFs.
I fooled around with it for a day or so and then did what I’ve done with a lot of projects. I put it to the side. It wasn’t that it was bad. It was just obvious that if I wanted to do anything serious with it, I’d have to commit real time. Learning how to assemble a KDP book properly, handling front matter, formatting, all of that. So it sat.
The site idea
Fast forward to around 2023.
I had started thinking again about building a Sudoku site. I ended up buying an expired domain, Sudoku Shark, thinking it might have some leftover authority or links. It didn’t really, but it wasn’t toxic either. I built out a bare-bones version of the site. Printables, the idea of selling books, the first rough shape of what eventually became Sudoku Depot.
That project stalled too. I stopped tinkering, let the domain go, and moved on to other things.
Picking it back up
Toward the end of 2024 or the beginning of 2025, I was looking through all my domains again, trying to decide what was worth pursuing. I remember having a conversation with ChatGPT where I basically laid everything out. What I knew how to do, what I enjoyed, what kind of time I realistically had.
There were a few ideas on the table. Print-on-demand was one of them. Living Dead Gamer lived there. That felt fun, low pressure. Sudoku Depot stood out differently. It felt more focused. More buildable.
The idea became simple: build a site around one puzzle type, sell printables, and publish books through KDP. Not ten puzzle types. Not everything under the sun. Just Sudoku.
Where things broke down
At first, I tried to use Puzzle Book Mastery exactly the way most people probably do. Generate PDFs. Create separate PDFs for title pages, tables of contents, thank you pages, and stitch it all together. I was using Canva to match fonts so everything looked coherent.
That’s when the friction showed up.
There were things I wanted on the puzzle pages that weren’t there. Small details. Information you’d expect to see in a book. Extracting puzzles, cutting them out, pasting them into new layouts quickly started to look like a mess. Slow, manual, and fragile.
I kept thinking, this shouldn’t take this long.
At some point, while talking through options, ChatGPT suggested something offhandedly. Unless you want to build your own generator.
That stopped me.
Build my own generator?
I asked if that was actually feasible. The answer was yes. Very feasible.
The bad decision that became the whole project
Looking back, this is the moment that feels like that scene in The Wolf of Wall Street where they ask Captain Ted if he can handle some chop. He pauses, does not want to turn them down, and says yes.
That was me. And ChatGPT was the captain.
Of course it said yes.
Suddenly, every option was on the table. Full control. Difficulty levels. Solving logic. Layout decisions. Everything sounded possible. In my head, I was picturing a polished software product with toggles and controls and a nice interface, the kind of thing you’d buy online.
That is not what I ended up building.
Frustration and false progress
This part took weeks. Probably months, if I’m honest. And it was miserable at times.
The generator wasn’t like coding a footer or a WordPress plugin. Those were done in minutes or a day. This was different. There were constant hiccups. Formatting issues. Indentation problems. Code being pasted into the wrong place. Tabs versus spaces breaking everything.
I kept asking for full drop-in files because partial snippets were a nightmare to merge. Long chats slowed down. Pages became unresponsive. We had to move to new chats just to keep things working.
The first time I almost quit for real was when I was told we were basically done. I was given a file to run. I opened it, and it was essentially nothing. Not even close to what we had been talking about for weeks.
I was burned out. Angry. I didn’t care about fonts or design anymore. I just wanted to print puzzles and move on.
What made it worse was the realization that ChatGPT wasn’t lying exactly, but it also wasn’t remembering. I had to go back and show earlier conversations where features were explicitly agreed on. That was the first time it really hit me that AI confidence and AI memory are not the same thing.
Switching tools
Eventually, I decided to try Claude and Cursor. People online kept saying that was the move for coding. And at first, it felt incredible. Things moved fast. It looked like the project might wrap up in days.
Then I hit new walls.
Claude would hit daily limits. Or it would add new code by silently removing old code. Or it would truncate sections and tell me to “refer back” to code that no longer existed. I had to explicitly tell it not to remove anything, not to summarize, not to truncate.
We also experimented with other puzzle types like Kakuro. Those were eventually scrapped. I narrowed everything down to Sudoku only.
By early 2025, the generator was probably 75 to 80 percent complete.
Then I abandoned it.
For months.
I changed my job location. I stopped touching the project. It sat.
Finishing it
Sometime around August, I picked it back up. We ran tests. Generated a full book. And for the first time, everything worked. Puzzles. Difficulty levels. Solutions. Layout. It was real.
Oddly enough, after finishing the hardest part, I stalled again. Covers became the new friction point. Then I stopped again for a while.
It wasn’t until late December that I forced myself to finish something tangible. The first book was generated and uploaded to KDP. The generator finally did the thing it was built to do.
What the generator actually is
One thing I misunderstood early on was what “building a generator” meant.
It’s not a shiny interface. It’s not a dashboard. It’s a script. I run it. It produces a complete, KDP-ready PDF. That’s it.
And honestly, that’s enough.
Looking back
This project took ChatGPT, Claude, Cursor, and a lot of patience. There were multiple points where I wanted to quit. Multiple times where I thought I had wasted months.
In the end, the generator exists. It works. And the hardest part is done.
Everything after this is execution.
