We use technology for almost everything, but what if we built something that encouraged us to step away from our screens?
That question inspired 🌱OuttaDoor, a quest-generation web application designed to turn ordinary moments into opportunities for real-world adventure.
🔗 Live Demo: https://outtadoor-1.onrender.com/
💻 Source Code: https://github.com/aditric7/OuttaDoor
🌱 The Idea Behind OuttaDoor
Sometimes, all we need is a small idea to break our routine: explore somewhere new, try an outdoor activity, or spend time doing something different.
OuttaDoor is designed to make that first step easier by generating quests that encourage users to engage with the world around them.
The idea fits the Hacktoberfest Week 1 theme, “Touch Grass,” by making the screen a starting point rather than the destination.
🛠️** How I Built it:--**
I built the application using:
1:HTML, CSS and JavaScript for the frontend
2:Python, FastAPI and Pydantic for the backend
3:Gemma 3 1B with Ollama for the local AI setup
4:Render for deployment
I worked on setting up the project structure, connecting the frontend and backend, configuring deployment, troubleshooting issues, updating documentation, and testing the live application.
I also used AI assistance during development. Along the way, I worked through the integration and deployment process to understand how the different parts fit together.
🤖 Why Open-Weight AI?
I explored running Gemma 3 1B locally through Ollama. This gave me an opportunity to learn about local inference and the flexibility of working with open-weight models.
A local model can offer more control over where inference happens and can reduce dependence on a remote model API when it is running on your own machine.
I also deployed the application on Render and implemented fallback quest generation so the website remains usable when the local AI service is unavailable.
There is an important limitation in my current deployment:
But the Render backend cannot automatically access the Ollama model running on my laptop. When that model is unavailable, the application falls back to predefined quests. So the live demo does not always perform AI inference.
✨ What Works Today?
1: A web interface for generating quests
2: A FastAPI backend connected to the frontend
3: A fallback mechanism for when the local AI service is unavailable
4: A deployed website that I tested independently of my laptop
🚀 ** What I'd Improve Next**
I'd like to make model inference available in the deployed version, improve quest personalization, and explore ways to make the experience more useful for different outdoor activities.
Building OuttaDoor has been a learning experience in turning an idea into a working, deployed application—not just writing code, but connecting components, debugging, and testing the result.
Try OuttaDoor: https://outtadoor-1.onrender.com/
*If you try it, I'd delighted to hear what you think! *🌿
Top comments (1)
tr.ee/dev-to