Sometimes, unexpected free time leads to unexpected adventures. After wrapping up my time at Social Seeder, I found myself with the rare opportunity to dive into something I’ve always wanted to explore: AI, storytelling, and product thinking. But how do you build an AI-powered storytelling engine when you’re not a programmer?

I’ll be blunt. Being jobless sucks. The endless cycle of applying, waiting (often hearing nothing), and dealing with uncertainty is exhausting. But amidst the frustration, I found something exciting. I finally had time to work on my motorcycle license (I got my test tomorrow, wish me luck! 🤩) and, more importantly, dive deep into product management and AI studies. More on that soon.

I’ll be blunt, being jobless sucks. You’re applying to jobs, waiting ages to hear something back (which is often nothing at all) and despite the fact that I’m applying a lot. It is stressful. I feel lucky I got the time to work on my motorcycle licence (I got my test tomorrow yay! 🤩) But despite that, I’ve spent most of my time at my product management and AI studies (more about that in the future). AI has empowered me so much in my hobbies. One of those areas is writing. After years of having a writers block, I started divig into roleplaying 6 months ago and it’s fun. I won’t talk about the expansive universe I’ve written so far, as I’m even playing with the idea of a childrens book.

AI Rekindled My Love for Storytelling

AI has empowered me in ways I never expected. After years of writer’s block, I started roleplaying six months ago, and it has been incredibly fun. I won’t go into the expansive universe I’ve built just yet, but I might even turn some ideas into a children’s book.

After diving deep into AI and prompt engineering, I wanted to take things further. I wanted to install my own AI model and actually see how it works. But the real question?

What kind of AI should I build? And how the F do you do that with zero real coding experience?

💡 The Challenge: Build an AI Dungeon Master With No Programming Experience

That’s when my friend ChatGPT and I came up with an idea.

Could I combine my love for history, storytelling, roleplaying, and technical curiosity into an AI-powered Dungeon Master?

How I Built MythosQuest (with AI Helping Me Learn AI 😆)

🚀 Step 1: Setting It Up – It Actually Works?!

At first, the idea sounded ridiculous. I don’t code. I’ve worked with Linux, product management, and AI prompt engineering, but building an AI from scratch? Where do you even start?

Step one was surprisingly easy. I set up Visual Studio Code, a WSL environment, and GitHub. Within an hour, ChatGPT had written a simple Python script that ran an AI model locally using Ollama and Mistral.

I was shocked. It actually worked. A self-hosted AI Dungeon Master with just a few lines of code. This was going great.

🚧 Step 2: Enter Dependency Hell – The ChromaDB Saga

But then came the real challenge. Making the AI historically accurate. I didn’t want King Louis XIV sending tweets instead of letters, so I needed a way to enhance its knowledge.

Enter Retrieval-Augmented Generation (RAG). ChatGPT recommended ChromaDB, a vector database to store historical references. Sounds easy, right? Nope.

What followed was five hours of debugging dependency hell. Python and SQLite refused to cooperate. After banging my head against the wall, I gave up and threw everything into a Docker container. I needed Docker anyway to run the project on an external VM.

Great, problem solved. Well… not really.

🐢 Step 3: The Battle with Docker and Why It’s So Slow

It took another ten hours of trial and error to get the right combination working.

Mistral refused to act like a Dungeon Master.
Docker kept rebuilding endlessly.
And I don’t even want to count how many times I swore at my screen.

Finally, after endless tweaking, it worked. I had a functioning AI Dungeon Master running in my terminal.

But there was one massive problem. It was SLOOOOOW.

I have a decent PC, but waiting five minutes per response was painful. My little Oracle server barely functioned, and my PC was overheating.

After a lot of experimenting, I reached a painful conclusion.

Running AI models locally is incredibly resource-intensive.

---

🔄 Pivoting to an API-Powered Dungeon Master

So, back to the drawing board. Instead of hosting everything myself, I decided to offload the gameplay to an AI API.

Since I have experience working with APIs, I chose CosmoRP, one of my favorite models for roleplaying.

The result?

✅ Significantly faster
✅ Way more responsive
✅ An actual usable AI Dungeon Master

What’s Next?

What started as a simple AI experiment has turned into a full-fledged project, and I’m just getting started.

✨ Expanding from basic choices to dynamic world-building
🧠 Improving memory, session recall, and multi-user support
📜 Refining historical accuracy while keeping gameplay fun

What started as an AI learning experiment turned into a deep dive into AI, infrastructure, and game design. And I’m having a blast.

💬 Curious about AI Dungeon Masters? Want to see how this experiment evolves?

👉 Check out the project on GitHub!