This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
What if an AI's most successful interaction with you was the moment you stopped talking to it?
That question became A Little Outside.
Most of the AI experiences we use are built around continuation: ask a question, get an answer, ask another question, and keep going.
I wanted to build something in the opposite direction.
A Little Outside is an AI experience designed to give you a small reason to leave the screen and experience the world around you.
You start by telling it how you're feeling or what's on your mind. You can also share a small personal memory, if you want, and tell it how much time you have.
The AI then turns that into a personalized outdoor moment.
Not a workout.
Not a productivity challenge.
Not therapy.
And not another endless conversation.
It might invite you to step onto a balcony, sit somewhere on campus, find a quiet corner of a garden, or simply walk somewhere nearby. From there, it may encourage you to notice the evening light, listen to the sounds around you, feel the air, watch the trees move, or find one small detail you normally wouldn't notice.
The important part is that it doesn't feel like a task.
It feels like an invitation.
A little memory can change the experience
I wanted the experience to be more personal than simply asking:
"What activity should I do?"
So I added an optional personal-memory layer.
A user might write:
"I used to sit outside at home in the evening and listen to the sounds around me."
The AI doesn't invent anything about that memory. It simply uses the feeling or atmosphere behind what the person shared to shape the experience.
And if someone doesn't have a memory they want to share, that's completely fine.
The memory is optional.
The feeling is enough.
The most important design principle
Throughout the project, I kept returning to one rule:
If a feature makes the user want to stay on the screen longer, it probably doesn't belong here.
That's why the final interaction is intentionally simple.
After receiving their outdoor moment, the user can choose:
🌿 I'm going outside
And the experience ends with:
Go.
You don't need to capture this moment.
You don't need to share it.
Just give yourself a few minutes to be there.
The screen is only the beginning.
The real experience happens outside.
Demo
🌿 Try A Little Outside
https://a-little-outside.onrender.com/
The application is publicly deployed and ready to try.
The experience follows a simple path:
What you're feeling
↓
Optional personal memory
↓
Available time
↓
Open-weight AI
↓
A personalized outdoor moment
↓
Optional music mood
↓
"I'm going outside"
↓
🌿
The goal isn't to make the AI conversation longer.
The goal is to make the real-world experience more meaningful.
Code
🌿 GitHub Repository
https://github.com/arnabduttacs478-droid/a-little-outside
The project is intentionally small and straightforward.
The core application is built with Python and Streamlit, with the AI interaction handled through Groq.
The repository also includes the instructions needed to run the project locally.
How I Built It
The project uses:
- Python
- Streamlit
- GPT-OSS-20B
- Groq
- GitHub
- Render
The AI layer uses GPT-OSS-20B, an open-weight model, through Groq.
The application gives the model three important pieces of context:
1. What the person is feeling
They can choose from simple starting points such as:
- "I need to breathe"
- "I miss home"
- "My mind is crowded"
- "It's been a difficult day"
- "I want a little happiness"
- "I just need to get outside"
Or they can write whatever is actually on their mind.
2. An optional personal memory
This allows the experience to connect with something familiar rather than generating a completely generic suggestion.
3. Available time
The user can choose between 5, 10, 20, or 30+ minutes.
The system prompt then guides the model to transform that information into a single outdoor experience.
A major part of the implementation was getting the tone right.
Early AI-generated suggestions could easily become cold or mechanical:
"Take a walk. Find a bench. Sit down. Observe your surroundings."
Technically, that works.
But it misses the point.
So the prompt was refined to make the AI behave more like a gentle companion than an instruction generator.
It encourages sensory details — sounds, air, sunlight, temperature, trees, surroundings — and uses flexible language such as "you could," "perhaps," and "if you'd like."
The model is also explicitly instructed not to:
- diagnose users,
- provide medical advice,
- make mental-health treatment claims,
- turn the experience into fitness training,
- invent details from personal memories,
- or turn the experience into a rigid checklist.
Music
Music is an optional part of the experience.
Instead of generating or reproducing copyrighted music, the AI can suggest a music mood or style, such as soft instrumental, gentle acoustic, or quiet piano.
Music is secondary.
The outside experience remains the point.
Deployment
The application is deployed publicly using Render.
The Groq API key is stored as a private environment variable rather than being committed to GitHub.
That keeps the public repository free of credentials while allowing the deployed application to access the AI model.
Why Does Open Innovation Matter?
For me, open innovation isn't just about using a different model.
It's about having the freedom to experiment with the part of the system that creates the experience.
A Little Outside isn't fundamentally tied to one AI model.
The application can potentially switch between different open-weight models while keeping the same core experience.
That opens possibilities for:
- experimenting with different models,
- comparing model sizes,
- fine-tuning,
- customizing model behavior,
- exploring local inference,
- self-hosting,
- and reducing dependence on a single closed provider.
This separation is particularly interesting for this project.
The AI isn't supposed to be the product.
The experience is.
The model is simply the component that helps turn someone's thoughts into a small invitation to go outside.
That means the underlying model can evolve without changing the fundamental idea.
There is also a deeper reason I wanted to approach the project this way.
AI is becoming extremely good at keeping us engaged.
It can answer almost anything.
It can continue almost any conversation.
It can give us another reason to look at the screen.
I wanted to experiment with something different:
AI that knows when its job is finished.
If the user closes the app and goes outside, that's not a failure of the product.
That's the success condition.
My Agent Session
I did not use DevRelay for this project, so I have left this optional section out.
Prize Categories
Best Use of Render
A Little Outside is publicly deployed using Render.
🌿 Live Demo
https://a-little-outside.onrender.com/
Sometimes we don't need another answer.
Sometimes we just need to look up.
At the sky.
At the trees.
At the street we've walked past a hundred times.
At a place that reminds us of home.
So I built an AI that gives you a little reason to do that.
A little AI. A little music. A little nature. A reason to step outside. 🌿
Top comments (0)