I was going through some old files recently - the usual ritual of deleting “final_v3_realfinal_updated(1).zip” — when I stumbled upon something I hadn’t seen in a while: the source code and pitch deck from my first startup.
It was called Fortune Folly, and we built it in 2023 at the peak of the ChatGPT hype, when half the world was trying to build something “AI-powered,” and the other half was trying to figure out what AI actually was.
The Birth of an Idea
In 2022, I was studying at the Innopolis University. Around that time, ChatGPT dropped and instantly became the thing everyone was talking about.
As students, we spent nights in the lab trying to break it, confuse it, or just chat about nonsense. It was fun, and honestly, a little magical. It didn’t feel like talking to a dumb bot anymore. The sentences were smooth, the logic made sense (most of the time), and you could almost forget there wasn’t a person on the other end.
But after the initial “wow” phase, one of my friends looked up from his laptop and said:
“What if ChatGPT could be a Dungeon Master?”
For those who aren’t familiar, DnD (Dungeons & Dragons) is a tabletop role-playing game where players act as heroes in a fantasy world, and one person acts as the Dungeon Master (DM). He plays everyone else: monsters, villagers, the weather, even the laws of physics.
So, the idea was simple but exciting: make ChatGPT the DM - an AI storyteller guiding a group of human players through an adventure.
Early Experiments (a.k.a. The Lab Campaigns)
Our first “sessions” didn’t even use the API. We literally played DnD in ChatGPT’s web interface from one account, typing things like:
Vladimir the wizard casts Fireball at the goblin.
Ilya the rogue hides behind a rock.
And ChatGPT would gamely respond with something like:
“The goblin screams in terror and drops its weapon!”
It was ridiculous and chaotic, but honestly, we were having a blast.
Of course, ChatGPT also had a habit of breaking character or suddenly deciding it “cannot simulate violence.”
That’s when we realised: if we wanted to make this a real multiplayer experience, we’d need to build something custom.
Building Fortune Folly
When OpenAI opened API access, we jumped in immediately. We built a Telegram bot that let multiple players chat with the same AI-controlled Dungeon Master.
We also wrote a backend service that handled dice rolls, story context, and message formatting, so ChatGPT could focus on narration instead of math.
It worked surprisingly well until it didn’t, because then we hit the Great Token Wall.
When Tokens Ruin Your Fantasy World
Every large language model like GPT reads and writes text in tokens- little pieces of words.
Roughly speaking, 1 token ≈ 4 characters.
The problem is: the model can only handle a fixed number of tokens at once - its “memory.”
For GPT-3.5 in 2023, that was around4096 tokens, or roughly 3000 words.
So what happens when your story goes beyond that?
You’d be fighting a dragon, and suddenly ChatGPT would act like it’s never met you before.
“Who is Vladimir the wizard?”
“Why are we in a cave?”
Imagine playing a DnD campaign with a Dungeon Master who gets amnesia every 5 minutes. That’s how our users felt.
The Thesis That Became a Startup
Luckily, our university had a program where you could turn your master’s thesis into a startup project.
It allowed us to spend more time on the project - code, pitch, brainstorm, and all.
We called it Fortune Folly, defended it in August 2023, and to our surprise got an A.
A few weeks later, we received the result from the Student Startup competition in Russia and actually won 1000000 rubles (about $10,000 at the time).
It felt surreal - we were not expecting this, but it gave us the opportunity to grow.
First Launch: The Folly of Fortune Folly
We polished the bot and web app, added some demo adventures, and opened it to players for testing. Around 1,000 people joined our early access.
And they loved it for about 5 minutes.
Then, as expected, the token limit kicked in and our AI Dungeon Master started forgetting everything again, so our sessions were limited.
So, I worked on a fix: a backend system that summarised past events, shortened the story, and fed the summary back into the model, kind of like giving ChatGPT a memory journal.
That extended sessions to about half an hour of consistent storytelling, which was a big improvement.
Our second test went much better.
We even opened our API for developers, allowing them to build their own worlds or narrative engines on top of Fortune Folly.
At that point, we realised - we weren’t just building a game; we were buildinga platform for AI-driven storytelling.
Reality Hits (and Sanctions Don’t Help)
Then came the hard part: turning it into a business.
Most of our users were from the US and Europe, but we were building from Russia, which, in 2023, was… let’s say not the easiest place to launch a startup. Only 30-40 users were coming from the local market.
Payment processing was a nightmare since the worldwide systems were not working, and local users weren’t that into text-based RPGs. There were no game developers interested in our product as well.
So we ran out of funding before we could expand.
We wrapped up the project, fulfilled all obligations for the student grant, and closed Fortune Folly.
Epilogue: Lessons from the Dungeon
I also found that one of our competitors - Hidden Doorreleased a similar idea: an AI-powered storytelling game in the 2024.
They had a beautiful interface, cards, and animations, but they didn’t includemulti-user sessions.
Looking back, Fortune Folly wasn’t a success story in the startup sense but it was one of the best learning experiences I’ve ever had.
We built something from scratch, worked nights and weekends, and created real magic with limited tools.
Even if no one remembers Fortune Folly, I’ll always remember the moment our AI Dungeon Master described a dragon attack with genuine drama before promptly forgetting who we were.