TouchGrass AI: Turning Screen Time into Real-World Adventures

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Submission 🌿

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What if AI didn't try to keep you on your screen longer, but helped you step away from it?

That's the idea behind TouchGrass AI — an AI-powered outdoor mission generator designed to help students, developers, and anyone who spends too much time in front of a screen get outside and enjoy the real world.

Instead of endlessly scrolling or wondering what to do during a break, users can choose how much time they have and what activity they prefer. TouchGrass AI creates a personalized outdoor mission with simple, practical steps.

The idea is simple: spend less time planning a break and more time actually taking one.

🌐 Live Frontend: https://touch-grass-ai.vercel.app

🔧 Backend: https://touchgrass-ai-zqxa.onrender.com

💻 GitHub Repository: https://github.com/Harsh2004-lgtm/TouchGrass-AI

The frontend and backend are deployed separately. The hosted AI inference integration is currently being stabilized; local development uses Ollama with Gemma 2B.

The complete project source code is available on GitHub:

👉 https://github.com/Harsh2004-lgtm/TouchGrass-AI

The repository contains the React frontend, Node.js/Express backend, database setup, and mission-generation logic.

During local development, I ran Gemma 2B through Ollama and connected it to my Express backend. The backend sends the selected activity and available time to the model, receives its generated mission, and saves the result in SQLite.

User selects time and activity ↓ React Frontend ↓ Express Backend ↓ Ollama + Gemma ↓ Outdoor Mission ↓ SQLite Storage ↓ Timer → Completion → Progress ↓ Streaks + Badges Enter fullscreen mode Exit fullscreen mode I also deployed the frontend through Vercel and the backend through Render. Getting the production AI inference provider working reliably has been one of the main deployment challenges.

For this project, open innovation is about having control over the AI powering the application.

Using an open-weight model with Ollama allowed me to experiment with AI locally, test prompts, and connect model inference to a real application without depending on a closed AI API during local development.

It also gives the project flexibility. The model and inference setup can be changed as the project evolves, without rebuilding the entire frontend and application flow.

There is a practical privacy benefit, too. In local mode, model inference runs on my own machine, so the mission-generation prompt does not need to be sent to a third-party AI inference service. The app can also be developed and tested without paying a per-request fee to a closed AI API, although hardware and hosting still have costs.

Most importantly, open-source AI made it possible for me to learn by building: from running a model locally to connecting it with an API, storing results, tracking progress, and deploying a full-stack application.

I wanted to use AI for something that encourages people to put their devices down. Open innovation gave me the freedom to experiment with that idea.

I want to make TouchGrass AI more reliable in production and continue improving the experience with features such as weather-aware missions, personalized outdoor recommendations, and more achievement milestones.

For now, the mission is straightforward: