*This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
I built TouchGrass AI, a local AI-powered outdoor adventure buddy designed to help people spend less time on screens and more time outside.
The user chooses an outdoor activity such as:
Hiking / Trekking
Walking
Gardening
Birdwatching
Running
Nature exploration and photography
The user provides their available time, experience level, general area, and preferences. TouchGrass AI then creates a practical outdoor plan with:
A simple timeline
Things to bring
Things to notice outside
An optional mini challenge
Activity-specific safety advice
A final reminder to put the phone away and go outside
The idea is simple:
Use AI to plan the adventure, then leave the screen behind.
Demo
The project runs locally on a Windows laptop using Ollama and an open-weight AI model.
Code
The complete source code is available here:
GitHub:
https://github.com/kuldeepdhanora40-prog/touchgrass-ai
The repository contains the Streamlit application, requirements, README, and project documentation.
How I Built It
I built TouchGrass AI using:
Python
Streamlit
Ollama
Qwen2.5 3B, an open-weight AI model
The architecture is:
User
↓
Streamlit UI
↓
Local Ollama API
↓
Qwen2.5 3B
↓
Personalized Outdoor Plan
↓
Put the phone away
↓
Go outside
Ollama runs locally on the user's computer. The application sends the activity request to the local Ollama API, where Qwen2.5 generates the outdoor plan.
I intentionally kept the application simple so that the AI is useful for a few seconds and then the user can leave the screen.
Why Does Open Innovation Matter?
Open innovation is important for TouchGrass AI because the project is designed around local, controllable AI rather than a closed AI API.
Using an open-weight model makes it possible to:
Keep data local
The user's preferences can remain on their own computer instead of being sent to a third-party AI service.
Swap models
The application architecture can use another compatible Ollama model without rebuilding the entire project.
Reduce API costs
After downloading the model, AI inference can run locally without paying for every generation.
Work without an internet connection
Once the application, dependencies, and model have been downloaded, the AI inference itself can run locally without an internet connection.
Experiment and customize
Because the model and local inference stack are accessible, developers can experiment with prompts, models, behavior, and application logic instead of being locked into one closed provider.
For this project, the open approach wasn't just a technical choice. It made local, private, customizable AI possible.
Touch Grass
The challenge theme inspired the main design decision of the project:
The screen should be the shortest part of the experience.
Instead of creating another AI application that encourages people to stay online, TouchGrass AI helps the user create a plan and then explicitly tells them to put the phone away.
The intended flow is:
Plan → Pack → Go outside → Touch grass.
My Agent Session
This project was developed with an AI-assisted coding workflow.
DevRelay Agent Session:
Add the DevRelay session link here if you have one.
The agent-assisted workflow helped turn the idea into a runnable local application and prepare the project for the challenge.
Prize Categories
Touch Grass
This project is specifically designed around the challenge theme: using open-source AI to encourage people to spend less time on screens and more time outdoors.
Some ideas I'd like to add in the future:
Offline trail and nature data
Local weather integration
Better hiking route planning
Bird-call identification
Garden planning based on local seasons
Outdoor challenges and streaks
More local open-weight models
For now, the goal is simple:
Don't let AI keep you at your desk. Let AI help you leave it.
Project
GitHub: https://github.com/kuldeepdhanora40-prog/touchgrass-ai
Top comments (0)