I’d like to share my experience building a web-based puzzle game, FRAC, developed entirely through prompt engineering and what I call “vibe coding”—using Gemini Flash 2.5 and GPT-4.1 as my primary coding partners.
My background is not in web development—I mainly work in data analysis and education—so this project was a personal experiment to see what could be achieved through persistent prompting and AI collaboration.
The idea was simple:
Could a non-developer, working step by step with large language models, build and deploy a complete online game?
I started with an idea for a logic-based puzzle game where players would solve number-based challenges, gradually uncovering a hidden story. Every aspect—from backend, frontend, and database to authentication and even the narrative itself—was developed with guidance from the AI.
There were many challenges along the way. I learned the basics of web frameworks, managed deployment and database migration, and encountered unexpected technical hurdles. In particular, when implementing OAuth authentication and deploying to Google Cloud Platform, I relied heavily on Gemini for guidance, troubleshooting, and understanding official documentation. In every case, the AI helped me understand the problems and suggest practical solutions. The process involved a lot of trial and error, and progress was sometimes slow, but it was also very rewarding.
The finished product is a mobile-friendly, ad-free game that is open for anyone to try. There are no monetization elements; the project was built entirely for the sake of exploration and learning.
If you’re interested, you can try the open beta here: fracgame.com.
No signup is required for casual play.
I’m sharing this mostly for others who might be curious about what’s possible with today’s LLMs, or who are considering a similar project. If you have any questions or thoughts, I’d be happy to discuss them.
