CoVerse

Inspiration

Beyond the classic tourist checklist, we want to form meaningful connections with the countries you chose to explore. Engaging on deeper levels with culture, traditions, and the people through interactive discovery challenges. When learning about a new country, most individuals concentrate on tourist attractions but are rarely given the opportunity to fully appreciate the country's traditions, language, and culture. We recognised that learning about cultures is often overlooked, yet it is what brings people together on a deeper level. This inspired us to build CoVerse, a platform that allows users to go beyond touring and immerse themselves in a culture through interactive, entertaining challenges. We intended to make learning a country's traditions and language not only easy but also enjoyable.

What it does

CoVerse is an interactive web platform that unlocks the world’s cultures through fun challenges.

Our main webpage features an interactive world map with pinpoints representing different countries. Users can click on any country to access its dedicated page and explore a range of cultural challenges. So far, we have created a Turkey Challenge page that includes the following:

Dialogue Challenge – Learn About Traditional Turkish Breakfast

Practice your conversational skills while discovering the ingredients, customs, and cultural roots of a traditional Turkish breakfast. This challenge allows users to pause and unpause audio from the chatbot and ensures the bot stops speaking when the user sends a voice message or text message, keeping interactions natural. The chatbot also highlights Turkish food-related words in each message to help users learn vocabulary in context. Users can send voice messages that are converted to text and seamlessly integrated into the chat, creating an immersive language learning experience. The use of audio playback enhances accessibility and supports efficient learning, while word-by-word translation helps users understand and remember new phrases more easily.

Turkish Idiom Generator

Discover Turkish idioms through AI-generated images and test your knowledge of Turkish expressions in a fun, visual way! Each challenge displays a Turkish idiom with its English translation. When the user presses a button, the Gemini image generation model (accessed through OpenRouter) creates a unique image representing that idiom. The user is then presented with multiple-choice answers and must select the correct meaning, making it an engaging MCQ-style quiz that blends creativity with cultural learning. Together, these experiences make learning about Turkish culture engaging, conversational, and visually inspiring.

Mind the Manners! – The Global Etiquette Game

“Think you know what’s polite? Think again.” This upcoming challenge is designed to test users’ knowledge of global etiquette. Players choose the right action in surprising cultural situations and learn how small gestures can have big meanings. While this challenge hasn’t been fully implemented yet, it is already displayed on the site as a preview of what’s to come.

How we built it

We built the web app using HTML, JavaScript, and CSS. Our challenges intergrated gemini api, specifically gemini-flash for text generations in our dialogue based challenges, and gemini-2.5-flash-image (Nano Banana) for image generation for our idiom challenge. Utilising Google AI Studio for gemini apis, as well as OpenRouter for gemini image generation models. Our dialogue based challenges are made into a more personalised learning experience through integrating ElevenLabs for Speech-to-Text features.

Challenges we ran into

Google AI Studio doesn't provide an image generation model in the free-tier so our alternative was using OpenRouter to access gemini-2.5-flash-image. Integrating ElevenLabs with gemini outputs. Merge conflicts in different branches as all our challenges were on the same web page rather than separate pages. Issues with api response payload format with different models. Matching interactive pin points to the map outline background. Adding more user-friendly features for more natural conversation style in our dialogue challenge. (Improving prompts given to gemini)

Accomplishments & Key Learnings we're proud of

Advanced Generative AI Integration (APIs) We successfully integrated multiple, generative AI services to build a rich, interactive user experience. ElevenLabs: Implemented a full-loop audio experience by integrating two separate endpoints:

Text-to-Speech (TTS): To dynamically convert the AI's chat responses into high-quality, natural-sounding audio, creating an immersive "virtual friend" (Beyza).
Speech-to-Text (STT): To enable user accessibility and convenience by allowing users to record their voice. We learned to capture browser-based audio, send it to the API, and process the transcribed text as user input.

OpenRouter Integration: We learned how to use OpenRouter as an intelligent API gateway. This powerful tool allowed us to access different AI models (Gemini, Claude, OpenAI, etc.).

Secure & Scalable Backend Architecture (Node.js) We learned the critical importance of server-side architecture for security and scalability, recognizing the risks of exposing API keys on the client-side. Node.js Server (as a Proxy): We designed and/or built a Node.js (likely using Express.js) backend to act as a secure proxy. This involved creating API endpoints (e.g., /api/chat) that the frontend could call. The Node.js server would then securely add the secret API keys (from .env files) and forward the request to the AI services, protecting our keys from being stolen.

Collaborative Development & Version Control (Git) Merge Conflict Resolution: We learned not just how to fix a merge conflict, but why they happen. This involved analysing conflicting code blocks (e.g., in script.js or index.html), communicating with teammates to understand their changes, and strategically merging different feature branches into the main branch without losing work.

What's next for CoVerse

Adding more unique culture challenges, and conversational practice unique to each country. Implementing a database for our idiom generation challenges with a dataset of most used idioms. Integrating LLM analysis of culture norms from country specific social media platforms.