AI Town

Inspiration

I have always been fascinated by how culture shapes the way people speak, think, and tell stories. At the same time, I am deeply interested in AI agents and human–AI interaction.

This project began with a simple question:

What if NPCs in a game could behave like culturally aware AI citizens instead of scripted characters?

I wanted to create a small world where virtual characters could represent different countries, speak with unique personalities, share real-world news from their homeland, and feel alive rather than pre-programmed.

This became Voices of the Town — a cultural exploration game powered by AI.

How I Built It

This project is a browser-based AI simulation built using modern web technologies, with Google’s Gemini 3 model serving as the intelligence core of the NPCs.

HTML/CSS for layout and visual design

Vanilla JavaScript (ES Modules) for game logic and interaction

Gemini 3 API as the main AI engine for NPC dialogue and personality

GNews API for country-specific real-world news

ElevenLabs Text-to-Speech for giving each NPC a unique voice

Gemini 3 as the Brain of the NPCs

At the heart of the project, Gemini 3 powers every conversation. Instead of relying on pre-written dialogue, each NPC generates responses dynamically based on:

Their cultural identity and personality prompt

The player’s message

Recent news from their home country

Each interaction is constructed through structured prompting, where Gemini 3 receives personality context and real-world information to produce natural, culturally grounded dialogue.

NPC Response=f(Personality Prompt,Player Message,Country News)

Gemini 3 enables the NPCs to:

Maintain a consistent personality

React differently to different players

Blend storytelling with real-world information

Feel less like scripted game characters and more like digital citizens

This transforms static NPCs into AI-driven agents capable of meaningful, context-aware conversation.

What I Learned

This project taught me a lot about AI integration in interactive systems:

Prompt Engineering for Personality

Small prompt adjustments can drastically change an AI character’s tone, politeness, humor, and storytelling style.

Real-World Data in Games

Integrating live news showed how external APIs can make virtual worlds feel connected to reality.

Voice as Identity

Adding Text-to-Speech made a huge difference. Voice adds emotion, cultural nuance, and a stronger sense of presence.

Multi-API System Design

I designed the system so multiple services (Gemini 3, News API, and TTS) worked together smoothly without blocking gameplay.

Challenges I Faced

Making NPCs Feel “Alive”

If prompts are too generic, characters feel robotic. If they are too detailed, responses become repetitive. Finding the right balance required extensive testing with Gemini 3 to keep personalities consistent but flexible.

Game Flow vs AI Latency

AI responses take time. I had to ensure the game still felt responsive while waiting for Gemini 3 text generation, news fetching, and voice synthesis.

Cultural Representation

It was important that NPCs felt culturally inspired without stereotyping. I focused on tone, interests, and storytelling style rather than clichés.

What This Project Represents

This is more than just a small browser game. It is an exploration of:

AI agents as digital citizens

Games as cultural bridges

Gemini-powered LLMs as tools for interactive storytelling

Voices of the Town shows how AI can transform static NPCs into living characters connected to the real world.