I built an AI that wants me to stop using it.

WHAT I BUILT

Okay so hear me out!!

We spend so much time staring at screens that we literally need apps to remind us that the world exists outside them and naturally, my response to this problem was to build another app.

Meet DETOUR.

DETOUR is an open-AI-powered adventure engine that gives you tiny real-world missions based on your mood, how much time you have, the weather, and the kind of adventure you're in the mood for.

You could have 10 minutes or two hours. You could be bored, curious, tired, restless, or just feeling a little chaotic. DETOUR takes that and turns it into an excuse to get out of your room and actually do something.

Not a workout plan. Not another productivity tracker telling you that you're behind in life and definitely not another app asking you to maintain a 47-day streak.

Just a tiny adventure.

For example, DETOUR might tell you to:

• Find something yellow that you would normally walk straight past.
• Take a different route home and notice three things you've never paid attention to.
• Find something that looks completely out of place, then walk another 200 steps.
• Go outside with a friend and let everyone pick a direction.

But there's more to it than just generating quests.

DETOUR gives your adventures a little progression system, with stats like Exploration, Curiosity, Observation, Courage, and Chaos. Apparently, even going outside needed character development.

Field Notes lets you record what you discover, write little reflections, and save your adventures.

The 1 KM Museum turns the ordinary things around you into a personal collection of discoveries.

And then there's Birdbrain, which helps you investigate the little things you encounter outside. Instead of immediately naming something, it gives you clues about what to observe.

The whole idea is to make the world around you feel a little less ordinary.

Because honestly, when was the last time you went outside without having a particular destination, deadline, or reason?

DETOUR is for people who want to explore their surroundings, people who get bored easily, people who need an excuse to leave their rooms, and anyone who wants to make an ordinary Tuesday slightly less ordinary.

The screen is not the product.

The adventure is.

DEMO

DETOUR is live.

You can go bother it here: https://detour-xqfa.onrender.com/

The app is hosted on Render's free plan, so it might need a little time to wake up if it has been inactive.

CODE

The code lives here: https://github.com/mehroofsaba/Detour

Feel free to explore the code, report bugs, suggest improvements, or contribute if something catches your eye.

HOW I BUILT IT

The tech stack is pretty straightforward, but getting the pieces to work together was the interesting part.

• Frontend: HTML, CSS, and JavaScript.
• Backend: Python and Flask.
• Database: SQLite.
• AI: Gemma through Ollama, with a procedural fallback when the model isn't available.
• Deployment: Render with Gunicorn.

I wanted the interface to feel like a little field journal rather than another generic AI dashboard. Think warm paper, botanical details, handwritten accents, and the sort of thing you'd want to keep open while planning your next tiny adventure.

The AI side was particularly important because I wanted to experiment with an open-weight model locally, while still making sure the deployed application worked when that model wasn't available.

DETOUR uses Gemma to generate missions that consider the user's available time, mood, weather, and other context. The prompt is designed to make the missions specific, a little strange, and safe. It avoids suggestions involving trespassing, driving, or unnecessary purchases, and rainy weather changes the kind of adventures it suggests.

I also built a procedural fallback so that the application can still generate missions when an Ollama model isn't available.

That mattered for deployment because Render's free web service doesn't run my local Ollama installation. The deployed app can use its fallback instead of pretending a local model is magically available in the cloud.

There is still plenty I'd like to improve, but I wanted a working application that people could actually try rather than a beautiful mockup with a very ambitious README.

WHY DOES OPEN INNOVATION MATTER?

This project is a big part of why I wanted to experiment with open-weight AI in the first place.

I didn't want AI to be just a mysterious API call hidden behind a button. I wanted to understand how the model fits into the application, how the prompts shape its behaviour, what happens when the model isn't available, and how much of the experience I can build and control myself.

Using Gemma through Ollama gave me a way to experiment with model-powered mission generation while keeping the rest of the application under my control.

And I think that's especially interesting for a project like DETOUR.

The point isn't to make people dependent on an AI that keeps them engaged with a screen. The point is to use AI to create a reason to leave that screen.

Open models make that kind of experimentation more accessible. You can build around them, inspect how your application uses them, experiment with prompts, and design fallbacks instead of treating a hosted model as the entire product.

For me, open innovation means being able to take an idea that sounds a little ridiculous, get my hands dirty, and actually build something with it.

Sometimes that something tells you to go outside.

PRIZE CATEGORIES

Open-Source AI / Touch Grass

DETOUR explores how an open-weight model can power real-world adventure generation. I use Gemma through Ollama for local mission generation, with a procedural fallback that keeps the deployed application usable when the model isn't available. The idea is to use AI to encourage people to explore the world around them instead of spending even more time on a screen.