Vibe City

Percy the pigeon, your guide through Vibe City
Pip the Pond Frog scene where players learn to refine prompts about water sounds
Sandy the Beach Crab scene teaching prompt precision by describing beach activities exactly
Vibe City home screen inviting players to help Percy the pigeon talk his way across the city

Inspiration Prompt engineering is becoming a critical skill in the AI era, but it's hard to learn. Most people don't understand that getting good outputs from AI isn't about asking longer questions, but rather it's about precision and technique. We wanted to create something that teaches this in a fun way. Percy the pigeon became our guide through this learning journey.

What it does Players guide Percy the pigeon through a day in the city by learning to communicate clearly with AI assistants (PIGEONBOT) to produce exact outputs. Each scene presents a mini-challenge where players must learn to remove layers of filler, including greetings, code blocks, and labels, to get the AI to output exactly what's needed. The game scores players on precision, speed, and efficiency.

How we built it We built Vibe City as a Next.js app with Turbopack, using Vercel and Claude for the AI interactions. Each story is file-driven (JSON-based), making it easy to create new challenges and stories. Through a Claude skill we designed, a developer user can create a new level by simply specifying the main subject they want to see in their level/story. The scoring engine uses embeddings and cosine similarity to measure how close the AI output is to the target, combined with a judge model (Haiku) for quality validation. We also use Claude for the main chatbot window. We come from non-technical backgrounds, so we learned a lot about full-stack deployment, and the Claude workshop helped us learn how to use subagents and use a skill to make the game generation scalable.

Challenges we ran into

Balancing difficulty across different problem types
Creating meaningful scoring that rewards both correctness and efficiency
Handling edge cases where embeddings might be misleading

Accomplishments that we're proud of

Seamless scene transitions with animation
A persona system that feels alive and consistent
A story, graphics, and problem creation are unified skills.
Teaching a complex concept (prompt refinement) through intuitive gameplay

What we learned Prompt engineering isn't about being clever, but rather it's about being specific. The best prompts are boring and precise. We learned that people learn better through iterative failure than through instruction.

What's next for Vibe City

More stories and characters beyond Percy
Multiplayer mode where players compete on the same challenges
Exposing our story creator tools to general users so players can design their own challenges
Mobile optimization