Touch Grass Planner: Local AI for Your Next Outdoor Reset

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

What I Built

I built Touch Grass Planner, a local AI app that helps people choose a real-world activity when they want to step away from the screen.

Instead of asking a closed API for recommendations, the app uses an open-weight sentence-transformer model running locally on the machine. The user enters a mood or goal such as “I want a calm walk with birds and fall colors,” and the app recommends a few outdoor options based on weather, time of day, energy level, and intent.

This is designed for people who want a small reset without planning a whole hiking trip: bird walks, garden sessions, sunrise strolls, social trail walks, and easy nature resets.

Demo

A live local version runs with Streamlit:

streamlit run app.py
Then open the local app in the browser at http://localhost:8501.

Code

GitHub repo: https://github.com/nirajnagrale7/hacktomberfest-second-challenge-2026.git

How I Built It

I built this with a simple local AI stack:

Streamlit for the app UI
Hugging Face sentence-transformers for the embedding model
PyTorch for on-device inference
A local JSON catalog of outdoor activities for matching user intent to real-world activities
The app uses the open-weight model to embed the user prompt and compare it against outdoor activity descriptions. It then ranks the best matches based on semantic similarity plus a few custom heuristics for weather, energy, time of day, and goal.

The important part is that the model runs on-device. There is no paid subscription, no remote API, and no external AI service required for the recommendation flow.

Why Does Open Innovation Matter?

Open innovation matters here because this project is built around privacy, affordability, and control.

The core experience works locally: the model runs on the user’s machine, the recommendation logic stays transparent, and the app can be customized or fine-tuned without being locked into a closed platform. That makes it more practical for real-world use in places where internet access may be unreliable or where a person wants to avoid sending personal preferences to a third-party server.

A closed model would have made this app harder to trust, harder to experiment with, and more expensive to operate. Using open-weight tools allowed me to build something that is lightweight, local, and adaptable — exactly the kind of setup that makes AI useful in everyday life instead of only in a cloud dashboard.

The project is intentionally small, but it shows how open-source AI can make an app more personal and more grounded in the real world.

My Agent Session

I used a local, open-source workflow to prototype the idea directly on-device and iterate on the recommendation logic without using a closed AI service.