This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
What if we used AI to help us spend less time on our screens?
For Hacktoberfest Week 1, I explored that idea through TouchGrass Quest Lab, a small experiment using Gemma to generate personalized outdoor quests.
The idea is simple: tell the model how much time you have, what kind of activity interests you, and what environment you're in. It generates a short mission that encourages you to step outside and notice the world around you.
Instead of suggesting another app to scroll through, the goal is to give you something to do that doesn't require looking at your phone.
I explored three scenarios:
The Concrete Jungle Safari: Observe tree branches, listen for natural sounds, and discover plants growing in unexpected places.
The Urban Leaf Detective: Compare leaf shapes, observe climbing plants, and notice the colors of flowers.
Garden Glimpse: Adapt plant-observation activities for someone with only five minutes and no desire to touch plants.
The most interesting part was the last scenario. When I asked Gemma to revise an existing quest for a more restrictive setting, it changed the activities and explained why it had made those changes.
This is currently a prompt-based experiment rather than a finished application. I wanted to investigate the model's ability to generate useful activities and respond to constraints before attempting a larger implementation.
Demo
The experiment was conducted using Gemma through Google AI Studio.
Demo artifacts: I'll include screenshots of the original quests and the revised Garden Glimpse response to show the actual model outputs.
Code
The experiment documentation, prompts, recorded outputs, and evaluation notes are available in my GitHub repository:
TouchGrass Quest Lab 🌿
A small prompt-based experiment with Gemma in Google AI Studio
Hacktoberfest 2026 — Week 1 theme: “Touch Grass”
TouchGrass Quest Lab explores whether an open-weight language model can generate short, safe, phone-free outdoor activities and revise them when the user's constraints change.
Scope note: This repository documents a manual experiment conducted through Gemma in Google AI Studio. It is not a standalone application, does not run inference locally, and does not claim offline operation. Check the challenge rules or ask the organizers whether a prompt-and-evaluation artifact meets the “build something” requirement before treating this as an eligible submission.
Research question
Can Gemma generate practical outdoor quests and adapt them when time, environment, or activity constraints change?
Setup and reproducibility
- Interface: Google AI Studio
- Model family: Gemma
- Inference: hosted through Google AI Studio
- Runtime, model version, generation settings, and date/time should be added by the author if…
The repository documents the prompts used, the observed model responses, and the limitations identified during testing.
How I Built It
I used Gemma through Google AI Studio and designed a series of structured prompts to explore three questions:
Can the model generate concrete outdoor activities rather than generic wellness advice?
Does it adapt its suggestions when the activity preference changes?
Can it revise a quest when the available time and environment become more restrictive?
Each prompt specified the activity preference, time available, setting, equipment constraints, and safety requirements.
I then reviewed the responses qualitatively, looking at relevance, actionability, constraint adherence, safety, and whether the activities could be completed without using a phone.
One limitation became clear: the model sometimes generated relevant activities without fully satisfying the request. For example, a gardening prompt produced plant-observation tasks rather than activities involving gardening itself. The revised five-minute quest also failed to explicitly repeat the time limit.
These small mismatches were useful findings. They reinforced the importance of evaluating AI-generated outputs instead of assuming that a plausible answer is necessarily a correct one.
Why Does Open Innovation Matter?
I chose Gemma because open-weight models offer developers opportunities to experiment with model behavior and, depending on the model's license and deployment setup, choose how inference is run or adapt a model to a particular task.
For a future version of TouchGrass Quest Lab, local inference could be interesting. It could potentially let users generate quests without sending their prompts to a remote inference service, and possibly support offline use once the required model and software are installed.
However, this experiment used hosted inference through Google AI Studio. I haven't tested local execution, offline functionality, fine-tuning, or a comparison against a closed model.
For me, open innovation matters because it makes those alternative approaches possible to explore. It also encourages developers to understand the trade-offs instead of treating an AI model as an opaque API.
I see TouchGrass Quest Lab as an early experiment in that direction: a small idea, a real model, observable outputs, and clear limitations.
My Agent Session
I used Google AI Studio for the experiment. I haven't saved a DevRelay agent session for this work, so there isn't a session link to share.
The prompts and recorded outputs are documented in the repository instead.
Prize Categories
Potential category: Gemma — Best Use of Gemma
Gemma was the model used in this experiment. However, given that this is currently a prompt-based evaluation rather than a complete application, I would verify eligibility with the challenge organizers before claiming this as a qualifying partner-category entry.
The goal of TouchGrass Quest Lab is straightforward: explore whether AI can help people spend a little less time on their phones and a little more time noticing what's around them.
Sometimes, the best interaction with technology is the one that encourages you to put it away.
Top comments (0)