This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass 🌿 OutsideIRL — Your world is bigger than your screen.
OutsideIRL is an outdoor mission companion designed to help students and people who spend too much time on screens reconnect with the real world.
Instead of recommending another app to spend hours in, OutsideIRL helps users discover small, achievable outdoor activities based on their available time, mood, and environment.
Whether you have five minutes between classes or an hour to explore your neighborhood, it helps you find a reason to step outside.
The idea is simple: make the screen the shortest part of the experience.
🌐 Live website: https://outsideirl.onrender.com Open the app, choose your preferences, generate a mission, and take it outside!
💻 GitHub repository: https://github.com/mohithreddydonala37/plant The project uses HTML, CSS, and vanilla JavaScript, without a frontend framework or build step.
I built OutsideIRL using HTML5, CSS3, and vanilla JavaScript.
The application includes a mission catalog, preference-based selection, mission rendering, completion tracking, and browser local storage.
I also designed an adapter for Gemma 3 through Ollama so the mission generator can use an open-weight model running locally. The model response is validated before being displayed, and the app has a curated fallback when local inference is unavailable.
The website is deployed on Render as a static site.
Current deployment limitation: the public Render website does not automatically connect to my computer's local Ollama server. Visitors use the curated fallback missions unless a separate remote inference service is configured.
Open innovation makes it possible to experiment with AI without making every interaction dependent on a proprietary API.
With an open-weight model such as Gemma 3, developers can explore local inference, inspect and adapt the application architecture, experiment with different models, and keep inference on their own machines.
For OutsideIRL, local inference offers an important potential benefit: users could generate personalized outdoor missions without sending their preferences to a third-party AI service. It also makes experimentation more accessible. A developer can build a prototype, modify the model integration, and explore different approaches without designing the entire product around one closed AI provider.
The distinction matters: the current public deployment uses curated fallback missions, while local Gemma 3 inference is available only when the local Ollama setup is configured and running.
Not included yet.
I'll add a DevRelay session link if I publish a session demonstrating the implementation process.
Open-Source AI Challenge — Touch Grass.