Draura

Inspiration

On our way to ShellHacks, a fatal crash caused chaos on Southbound 65 reminding us how even skilled drivers face unpredictable, dangerous situations. Yet, many drivers go years without meaningful training. Standard driving test preparations rarely prepare people for emergencies like pedestrians running into traffic, scooters flipping over, or broken signals after disasters. Draura makes testing and practice interactive, locally relevant, and attuned to real world hazards.

What it does

Draura is an immersive 3D driving simulation that combines real Google Maps data with retro Game Boy looks. Players drive through real city streets while answering real-world situational driving questions, earning points for staying on roads and losing points for off-road driving. The game features authentic traffic scenarios, real-time road detection, and a nostalgic pixelated interface that makes learning engaging and fun.

How we built it

Built frontend with Three.js, Google Maps API, and TypeScript. Built backend with Python, FastAPI, Google Cloud Platform, Vertex AI, Gemini AI, Google Agent Development Kit, Google Cloud Core, Poetry, and Uvicorn for scalable backend logic. The game uses Auth0 for user authentication.

Challenges we ran into

Implementing accurate road detection required exploring multiple methods before using google maps road API with extensive error handling
Creating a scalable question system that could easily expand to cover more driving scenarios while maintaining consistent gameplay mechanics presented design challenges
Choosing the right type of Agent was difficult, as we had to dive deeper on how we would formulate a Driving Question by assessing past transportation related incidents and the common traffic signs in a given area from the user.
Deploying the agent was the time consuming part where configuring the right dependencies, and adjusting the environment for deployment

Accomplishments that we're proud of

Successfully created a seamless blend of real-world mapping data with engaging gameplay mechanics
Developed a robust road and road objects detection system that works with actual street layouts
Built a scalable multi AI-agent question system that can easily expand to cover more driving scenarios

What we learned

Learning how to use multi AI agent systems for the backend architecture taught us the importance of distributed processing and intelligent task delegation
Game development using Three.js and TypeScript revealed the power of type safety in complex 3D applications and the need for careful memory management in real-time rendering
Educational games benefit from immediate feedback and point systems that reinforce learning objectives
Performance optimization is crucial when combining multiple APIs and 3D rendering, especially when coordinating between AI agents and frontend visualization systems

What's next for Draura

Expanding to cover all 50 states' driving laws, adding multiplayer functionality for competitive learning, implementing voice recognition for hands-free question answering, and developing mobile versions for broader accessibility. We're also exploring integration with actual driving schools and DMV preparation programs.