🌿 TrailMate AI — Your Offline AI Outdoor Companion
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
TrailMate AI is an offline-first AI outdoor companion designed to help people spend less time looking at screens and more time exploring the real world.
The project uses Gemma, an open-weight AI model, running locally through Ollama as a core part of the application.
The idea is simple:
Discover → Ask Gemma → Get a mission → Put the phone away → Explore → Return → Record your discovery → Earn XP.
TrailMate allows users to take or upload a photo of a plant or natural object. The local AI system uses Gemma to analyze the observation and provide useful, beginner-friendly information.
Instead of keeping the user inside the application, TrailMate then uses Gemma to generate a short outdoor mission.
For example:
🌿 Outdoor Mission
Find another tree nearby and compare its leaves with the plant you just discovered.
+20 Explorer XP
The user can activate Touch Grass Mode, put their phone away, and complete the mission outside.
After returning, they can record what they discovered in their Nature Journal and earn Explorer XP.
Who is it for?
TrailMate is designed for:
- 🌱 Students
- 🥾 Hikers and walkers
- 🌳 Nature lovers
- 🐦 Outdoor explorers
- 👨👩👧 Families
- Anyone who wants a reason to spend more time outdoors
The goal isn't to make people spend more time using an AI app.
The goal is to make them stop using it and go outside.
🤖 How Gemma Powers TrailMate
Gemma is the AI brain of TrailMate.
Instead of sending every request to a proprietary cloud AI API, TrailMate runs a Gemma model locally through Ollama.
The basic architecture is:
User
│
↓
TrailMate Streamlit App
│
↓
Python Application
│
↓
Ollama
│
↓
Gemma Open-Weight Model
│
├── Nature/Image Analysis
├── Plant Information
├── Outdoor Mission Generation
├── Observation Feedback
└── Nature Journal Assistance
Gemma is therefore not just an extra chatbot inside the project.
Gemma is part of the core workflow that makes TrailMate work.
🌿 What Gemma Does
1. Nature Understanding
When the user provides a nature observation, Gemma helps analyze and explain it.
Example:
User:
I found this plant during my walk.
↓
TrailMate
↓
Ollama
↓
Gemma
↓
Possible identification
+ Description
+ Habitat
+ Interesting information
The application is designed to communicate uncertainty rather than pretending that every identification is correct.
2. Outdoor Mission Generation
Gemma converts an observation into an outdoor activity.
For example:
Observation:
User found a Neem-like tree.
↓
Gemma
↓
🌳 Mission:
Find another tree nearby.
Compare its leaf shape and size
with the first tree.
Reward: +20 XP
This is an important part of the Touch Grass concept.
3. Nature Journal
The user can describe what they saw.
Gemma can turn the observation into a short, useful reflection.
Example:
User:
I found two trees with different leaves.
↓
Gemma
↓
Nature Journal:
"Today you noticed two different
leaf shapes. Next time, compare
their bark and arrangement."
4. AI Feedback
After completing a mission, Gemma can provide feedback and suggest what the user could observe next.
This creates a loop:
Discover
↓
Gemma
↓
Explore
↓
Observe
↓
Gemma
↓
Learn
🛠️ How I Built It
TrailMate AI is built using:
| Technology | Purpose |
|---|---|
| Gemma | Open-weight AI reasoning/analysis |
| Ollama | Runs Gemma locally |
| Python | Application logic |
| Streamlit | User interface |
| Pillow | Image processing |
| SQLite | Local data storage |
| Git/GitHub | Open-source development |
Core technology stack
🌿 TrailMate AI
│
├── Streamlit
│
├── Python
│
├── Ollama
│ │
│ └── 🤖 Gemma
│
├── Pillow
│
└── SQLite
📴 Local AI Architecture
One of the key design decisions was to run Gemma locally.
The application communicates with Ollama on the user's computer:
TrailMate
│
↓
localhost
│
↓
Ollama
│
↓
Gemma
│
↓
Response
This means the core AI interaction does not need to depend on a remote proprietary AI API.
Once the required model and dependencies are installed, the application can continue to provide its core AI functionality without continuously requiring internet access.
🔒 Why Does Open Innovation Matter?
Open innovation matters because TrailMate deals with personal observations, images, and outdoor exploration.
Using an open-weight model such as Gemma gives us much more control over where and how AI runs.
🔒 Privacy
With local Gemma inference, the user's images and observations can remain on their own computer instead of automatically being sent to a proprietary AI provider.
📴 Offline Potential
Outdoor environments don't always have reliable internet.
Running Gemma locally through Ollama allows the AI experience to continue working after the model has been downloaded and configured.
🔄 Model Freedom
The architecture isn't permanently tied to one proprietary AI API.
We can experiment with different open models and choose the one that works best for specific tasks.
💰 No Per-Request Cloud AI Cost
Because Gemma runs locally, the core AI functionality doesn't require paying a cloud provider for every AI request.
This makes experimentation much more accessible for students and developers.
🛠️ More Control
Using an open-weight model allows developers to experiment with:
- Prompts
- Model selection
- AI behavior
- Local inference
- Custom datasets
- Future fine-tuning
- Specialized nature applications
This would be much harder to control when the entire AI layer is a closed API.
🌳 Why Gemma Was a Good Fit
We chose Gemma because it gives TrailMate a practical way to experiment with an open-weight AI model while keeping the AI layer under our control.
Instead of building:
TrailMate
↓
Closed AI API
↓
Internet
↓
Cloud
we built the core AI path around:
TrailMate
↓
Ollama
↓
Gemma
↓
Local AI
This architecture fits the project's goal of creating an offline-first outdoor companion.
🌿 The "Touch Grass" Philosophy
The most important design decision in TrailMate is that the AI should not become another reason to stay on the screen.
Many AI applications encourage users to keep chatting.
TrailMate does the opposite.
The application gives the user a short interaction:
📷 Discover something
↓
🤖 Gemma explains it
↓
🥾 Gemma creates a mission
↓
📱 Put the phone away
↓
🌳 Go outside
↓
👀 Explore
↓
📔 Record what you discovered
↓
🏆 Earn XP
The screen is only a tool to start and finish the experience.
The real product is the outdoor experience.
🥾 Touch Grass Mode
TrailMate includes a dedicated Touch Grass Mode.
When the user starts a mission, the application displays:
🌳 MISSION STARTED
Your AI guide has given you
everything you need.
📱 PUT THE PHONE DOWN.
🥾 GO OUTSIDE.
👀 LOOK AROUND.
🌿 EXPLORE.
Come back when you're done.
This is intentionally designed to minimize screen time.
🏆 Explorer XP
TrailMate includes a simple gamification system.
Examples:
🌿 Identify a plant +10 XP
🥾 Complete a mission +20 XP
📔 Add a journal entry +10 XP
🐦 Observe a bird +15 XP
🌳 Make a new discovery +25 XP
Explorer levels:
🌱 Beginner Explorer
↓
🌿 Nature Explorer
↓
🌳 Trail Explorer
↓
🥾 Outdoor Adventurer
↓
🏆 Nature Master
The purpose of XP is not to encourage more screen time.
It is to reward real-world exploration.
📔 Nature Journal
Every completed exploration can become part of the user's local Nature Journal.
The journal can contain:
- Date
- Discovery
- Image
- AI identification
- User observation
- Mission
- XP earned
Example:
October 7, 2026
🌿 Discovery:
Neem-like tree
👀 Observation:
"I found this tree during my walk."
🤖 Gemma's feedback:
"Compare the leaf arrangement with
another nearby tree."
🏆 XP:
+20
🔐 Privacy by Design
TrailMate is designed around local-first data handling.
User data such as:
- Nature photographs
- Observations
- Journal entries
- Mission history
- XP
can be stored locally.
The application does not need to send these details to a cloud AI service for the core workflow.
💻 Code
GitHub Repository:
[Add your GitHub repository link here]
The repository contains:
TrailMate-AI/
│
├── app.py
├── pages/
├── core/
├── database/
├── utils/
├── data/
├── requirements.txt
├── README.md
└── LICENSE
The project is intended to remain open so other developers can experiment with the AI prompts, model, outdoor missions, and future nature-focused features.
🎥 Demo
Demo Video:
[Add your YouTube/Loom/demo link here]
Live Demo:
[Add your deployed Streamlit link here]
The demo will show:
- TrailMate dashboard
- Gemma/Ollama local AI status
- Nature image upload
- Gemma-powered analysis
- Outdoor mission generation
- Touch Grass Mode
- Mission completion
- Nature Journal
- Explorer XP
🤖 My Agent Session
DevRelay Agent Session:
[Add your DevRelay session link here]
The session shows the development process and how the project was built with AI-assisted development.
🏆 Prize Categories
🌿 Touch Grass
TrailMate is designed specifically around the Touch Grass theme by encouraging users to leave the screen and explore their physical surroundings.
🤖 Open-Source / Open-Weight AI
TrailMate uses Gemma, an open-weight AI model, as a core part of the application.
Gemma runs locally through Ollama, allowing the project to experiment with local AI rather than depending entirely on a proprietary cloud API.
📴 Local / Offline AI
The architecture is designed around local inference, allowing the core AI workflow to work without continuously requiring internet access after the required model and dependencies are installed.
🚀 What's Next?
Future versions could add:
- 🐦 Offline bird-call identification
- 🌱 More specialized plant recognition
- 🗺️ Nature trail recommendations
- 🌦️ Weather-aware missions
- 🎙️ Voice interaction with Gemma
- 📍 Local biodiversity information
- 📴 Better offline maps
- 🏆 Group outdoor challenges
- 🧠 Fine-tuning with local nature datasets
The long-term goal is to make TrailMate a personal open AI guide for exploring the physical world.
💡 Final Thought
AI doesn't always need to keep us in front of a computer.
Sometimes the best thing an AI can do is tell us:
"You've learned enough. Put the phone down and go see it yourself." 🌳
That's what TrailMate AI is trying to build.
👤 Creator
Project: TrailMate AI
Challenge: Hacktoberfest Open-Source AI Challenge — Week 1: Touch Grass
Creator: Root Kevadiya
Top comments (0)