NatureQuest AI: Less Screen, More Sky — A Local Gemma-Powered Outdoor Companion

Sameer Akhtari — Sat, 10 Oct 2026 22:52:25 +0000

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

What I Built

What if the best AI application was the one that encouraged you to close your laptop?

We spend so much time building technology that keeps people engaged, clicking, scrolling, and staring at screens. For Hacktoberfest 2026, I wanted to explore the opposite idea: Could AI encourage us to step away from technology and reconnect with the world around us?

That question led me to build NatureQuest AI, an offline-first outdoor exploration companion powered by Google's open-weight Gemma 3 model.

NatureQuest transforms ordinary outdoor experiences into small, personalized adventures. Instead of spending hours interacting with an AI chatbot, users receive a few meaningful activities that encourage them to put their devices away, explore their surroundings, and notice details they might otherwise overlook.

How NatureQuest works

The application revolves around four main features.

1. AI-Powered Outdoor Adventures

Users choose three things:

Environment: Park, backyard, city streets, hiking trail, or coastline.
Mood: Curious, calm, active, or creative.
Available time: From a quick 10-minute break to a longer outdoor adventure.

Using these preferences, Gemma 3 generates three personalized outdoor activities.

For example, a curious user exploring a city might be encouraged to observe different leaf shapes, look for plants growing between buildings, or discover interesting natural patterns.

The idea is simple: give people a reason to go outside, not another reason to stay online.

2. AI-Powered Nature Observations

After completing an adventure, users can upload a photograph of something interesting they discovered.

NatureQuest uses the vision capabilities of Gemma 3 4B, running locally through Ollama, to examine the image and provide:

A description of visible details.
Interesting observations about the subject.
Suggestions for what to explore or notice next.
An explanation of the AI model's uncertainty.
Safety guidance for observing nature responsibly.

The model is deliberately instructed not to make unsafe assumptions about species identification, poisonous plants, or whether something is safe to eat or touch.

3. Personal Field Journal

Every adventure deserves a little memory.

NatureQuest includes a personal field journal where users can save completed activities, photographs, AI observations, and their own notes.

The journal uses the browser's IndexedDB storage, meaning entries remain on the user's device rather than being uploaded to a cloud database.

Users can revisit their discoveries, remove entries, and export their journal as JSON.

There are no leaderboards, unnecessary achievements, or endless notifications. The experience is about personal curiosity rather than competition.

4. Offline-Friendly Exploration

Nature shouldn't require an internet connection.

When the local AI model is unavailable, NatureQuest automatically switches to curated outdoor activities so users can continue exploring without waiting for inference.

The application also uses Progressive Web App (PWA) capabilities to cache the interface for subsequent offline visits.

AI-powered image analysis still requires access to the configured Ollama server, but outdoor missions and previously saved journal entries can remain available without the AI.

Bringing the idea into the real world

One of my observation examples involved a photograph of the night sky, where stars, cloud cover, and the surrounding environment created an interesting scene.

The image was saved in NatureQuest's field journal as "Night Sky with Cloud Cover", alongside an AI observation describing visible conditions and suggesting additional details to watch for.

I also saved a completed "Urban Observation" adventure containing three outdoor exploration activities.

These examples demonstrate the kind of experience I wanted to create: using a small amount of technology to encourage genuine curiosity about our surroundings.

NatureQuest isn't meant to replace the outdoor experience. It's meant to inspire it.

Demo

Here's my recorded demonstration of NatureQuest AI, showcasing the application and its outdoor exploration features.

Watch on YouTube: https://youtu.be/JD7JUCscFDk

The project demonstrates how a locally hosted open-weight AI model can be integrated into an outdoor exploration application with personalized adventures, image-based observations, and a private field journal.

Code

GitHub Repository:

https://github.com/sameerakhtari/naturequest-ai

sameerakhtari / naturequest-ai

Offline-first AI-powered outdoor exploration assistant — Hacktoberfest 2026

🌿 NatureQuest AI

Less screen. More sky.

NatureQuest is an offline-first outdoor exploration companion built for Hacktoberfest 2026 — Week 1: Touch Grass.

Plan a three-step outdoor micro-adventure, put your phone down, notice something wonderful, and return with a photograph. Google's open-weight Gemma 3 vision model runs through your own Ollama server and helps describe the details. A private field journal stores discoveries in your own browser.

▶ Watch the NatureQuest AI demo · Official challenge · Gemma 3 model · Ollama

Features

Adventure planner: Choose park, backyard, city, trail or coast; then a mood and duration. Locally running Gemma creates three outdoor prompts. When the model is unavailable, the app visibly uses a curated offline mission instead.
Nature observation: Upload or photograph a JPEG, PNG, or WebP image under 8 MB. The local server strips metadata, resizes it, and passes it to Gemma 3 vision for a description…

View on GitHub

NatureQuest AI is an open-source project released under the MIT License.

The repository contains the frontend, backend, AI integration, Docker configuration, documentation, and automated tests.

Main technologies

Frontend — HTML, CSS, and Vanilla JavaScript

I deliberately avoided a heavy frontend framework. The interface is responsive, mobile-friendly, and designed around simplicity, with a nature-inspired visual style.

The frontend handles adventure planning, photo uploads, AI responses, and the personal journal.

AI Model — Google Gemma 3 4B

Gemma 3 is the foundation of the application's AI features.

Its text-generation capabilities create personalized outdoor missions, while its multimodal vision capabilities help analyze photographs collected during exploration.

Local AI Runtime — Ollama

Ollama provides the local inference runtime.

Instead of sending images and prompts to a proprietary cloud AI service, NatureQuest communicates with an Ollama instance controlled by the application operator.

Backend — Python, FastAPI, and Pillow

FastAPI provides the endpoints connecting the frontend to Ollama.

Pillow handles image preprocessing, including file validation, resizing, and removing metadata during image conversion before inference.

Local Storage — IndexedDB

IndexedDB stores photographs, personal notes, and completed activities within the browser.

No external database or user account is required.

Offline Support — Progressive Web App

A service worker caches the application's static assets, allowing users to revisit the interface and use curated adventures without an active internet connection after the initial installation.

Deployment — Python or Docker Compose

The application can be started directly using Python or deployed through Docker Compose.

It is designed to be self-hosted without requiring commercial AI API keys or paid hosting infrastructure.

How I Built It

My main technical challenge was designing an application where AI serves the real-world experience instead of becoming the experience itself.

I wanted local inference, a lightweight interface, and useful functionality even when the model wasn't accessible.

1. Running an open-weight vision model locally

I chose Gemma 3 4B, Google's open-weight multimodal model, and integrated it with Ollama.

The application's backend communicates with Ollama through its local chat API.

This allows the same model to handle two different tasks: generating creative outdoor activities and analyzing images.

The inference endpoint is configurable, so users can run the application on their own computer or connect it to an Ollama server within infrastructure they control.

No proprietary hosted AI API is required.

2. Generating structured outdoor adventures

For adventure creation, the frontend sends the selected environment, mood, and duration to the FastAPI backend.

The backend builds a prompt containing these preferences and asks Gemma to generate a structured JSON response.

The expected response contains:

An adventure title and short description.
Three outdoor observation activities.
A closing message encouraging the user to put their device away.

The backend validates the AI response before returning it to the frontend.

If the model is unavailable or produces an invalid response, the application falls back to a predefined collection of outdoor activities.

Importantly, the interface distinguishes between Gemma-generated and curated adventures.

This means the application remains useful even when AI inference is unavailable.

3. Adding multimodal image understanding

The next challenge was giving users a way to learn from photographs collected during their adventures.

I developed an image-processing workflow where the user uploads a photograph through the frontend.

The FastAPI backend validates the uploaded file, checks size limits, processes it using Pillow, and removes embedded EXIF metadata during conversion.

The processed image is resized and transmitted to the locally configured Ollama endpoint for analysis by Gemma 3.

The AI returns a structured observation that the application displays in a readable format.

I also included explicit uncertainty and safety requirements in the model prompt, because image recognition models can make mistakes, particularly when identifying natural objects or living organisms.

4. Creating a private, persistent field journal

I wanted users to preserve their discoveries without creating an account or depending on a hosted database.

IndexedDB was a natural fit.

The frontend can save completed adventures, compressed photographs, optional notes, and AI-generated observations directly in browser storage.

Entries remain available after refreshing the page, provided the browser's site data is retained.

Users can also export their journal to a JSON file for backup.

5. Designing for less screen time

The user experience was just as important as the AI integration.

I designed NatureQuest around a few deliberate principles:

Generate short adventures instead of endless content.
Keep navigation simple and mobile-friendly.
Encourage outdoor exploration rather than extended interaction.
Avoid tracking, unnecessary accounts, and notifications.
Allow meaningful functionality when AI is unavailable.

The visual design uses nature-inspired colors, original illustrations, and a clean layout to make the application feel like an invitation to explore.

6. Testing and documentation

The repository includes automated API tests covering mission generation, offline fallback behavior, request validation, image handling, and mocked model responses.

It also includes setup instructions, Docker Compose configuration, a testing checklist, and a documented architecture.

Development was AI-assisted, with the source code and implementation details available publicly for inspection and improvement.

The automated tests use mocked Ollama responses; real-model behavior and browser interactions require separate hands-on verification.

Why Does Open Innovation Matter?

For me, this is the most important part of NatureQuest AI.

Most modern AI applications rely on remote APIs. While those services are convenient, they can introduce costs, infrastructure dependencies, and privacy considerations.

NatureQuest takes a different approach.

By using an open-weight model and local inference, I could build an AI application where users have more control over their data and the technology powering the experience.

Privacy without requiring a cloud AI service

Outdoor photographs can contain more information than we realize.

A photograph might include recognizable locations, buildings, people, or other personal details.

With NatureQuest's default local setup, image inference happens through Ollama on infrastructure controlled by the operator.

Photographs are not automatically sent to a third-party hosted AI inference provider.

The personal journal is also stored locally within the browser.

This architecture made privacy an important part of the product design rather than something added afterward.

Freedom to experiment and improve

Using Gemma 3 locally means the AI component isn't tied to a single proprietary API service.

Developers can inspect the prompts, adjust the application's behavior, experiment with different models, and improve the response validation.

This flexibility is especially valuable for projects where responsible AI behavior matters.

For example, I could specifically design the observation prompts to emphasize uncertainty and discourage potentially unsafe identification claims.

More accessible experimentation

NatureQuest can run on personal hardware without requiring a paid AI API subscription.

There are still hardware, storage, and electricity costs, and local inference can be slower than cloud services. However, the absence of per-request API charges makes experimentation more accessible.

The project also demonstrates that a relatively simple stack—Python, FastAPI, Ollama, and vanilla JavaScript—can support a useful multimodal AI application.

Technology that encourages real-world connection

Most importantly, open innovation made it possible to experiment with a different philosophy of AI.

Instead of building technology that demands constant attention, I wanted to create something that provides a useful suggestion and then encourages people to step away.

An AI system doesn't always need to maximize engagement.

Sometimes, the most meaningful interaction is the one that ends with someone closing their laptop, stepping outside, and discovering something they hadn't noticed before.

That's what NatureQuest AI is about.

Prize Categories

Best Use of Gemma

I'm entering the Best Use of Gemma category because Google's Gemma 3 model is fundamental to NatureQuest AI.

Gemma is responsible for two central features:

Personalized adventure generation: Creating outdoor exploration activities based on a user's environment, mood, and available time.
Multimodal nature observations: Analyzing outdoor photographs and providing grounded observations, follow-up suggestions, and uncertainty information.

The ability to run Gemma locally through Ollama also directly supports NatureQuest's emphasis on privacy, self-hosting, and user control.

Without the open-weight model, the project would lose a central part of both its functionality and design philosophy.

Final Thoughts

NatureQuest AI started with a simple question:

What if AI could help us spend less time looking at screens and more time appreciating the world around us?

The result is an outdoor exploration companion that combines local AI, personalized adventures, multimodal observations, and a private field journal.

There is plenty of room to improve it, including more outdoor activities, richer nature observations, and support for additional local AI models.

But the underlying idea will remain the same.

Technology should help us experience life, not distract us from it.

NatureQuest AI 🌿 — Less Screen. More Sky.

Thanks for checking out my Hacktoberfest 2026 project!

DEV Community: Sameer Akhtari