WildQuest AI

Akash Guha Thakurata — Sun, 11 Oct 2026 14:18:24 +0000

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

WildQuest AI 🌿 — Explore More, Scroll Less

📵 The Reality Check

The average developer spends 8 to 11 hours a day staring at glowing pixels. Most modern apps are engineered as slot machines—maximizing session lengths, retention graphs, and dopamine loops.

WildQuest AI was built to do the exact opposite. It is an AI companion designed to put itself away.

🧭 What I Built

WildQuest AI is a mobile-first, offline-first outdoor adventure companion powered by Google’s open-weight Gemma 2B model. Instead of keeping your eyes fixed on a screen, it generates personalized micro-adventures that invite you to step into your backyard, park, or trail to notice the living world with your own eyes and ears.

Feature	The Modern Screen Trap	WildQuest AI 🌿
Primary Goal	Maximize time spent in-app	Get you out of the app within 30 seconds
Interaction	Endless algorithmic feeds	One sensory prompt at a time
Hardware	Constant battery drain & notifications	High-contrast "Screen-Light" trail mode
Privacy & Data	Cloud tracking, ad cookies & location telemetry	100% on-device (IndexedDB + Local Gemma)
Connectivity	Fails completely without cell reception	100% offline-ready with procedural fallback

🌿 Quests That Teach You to See Again

Users pick their immediate environment (backyard, urban street, park, trail, waterfront), their time window (5 to 60 mins), and their curiosities:

🍂 Botanical Patterns: "Find three distinct leaf shapes on a single branch without picking them."
🐦 Acoustic Mindfulness: "Close your eyes for two minutes. How many distinct bird calls can you isolate?"
🎨 Color Perception: "Search for five different shades of green within ten steps of where you stand."
🐝 Pollinator Watch: "Observe bees or hoverflies on a blossom from two feet away without disturbing them."

🛑 The Anti-Screen Philosophy

When you hit Start Exploring, the interface shifts into Adventure Mode with a single instruction:

╔══════════════════════════════════════════════════════════════════════╗
║  "You're ready. Look up. Put your phone in your pocket and observe." ║
╚══════════════════════════════════════════════════════════════════════╝

No feeds. No badges for streaks you missed. Just an unhurried invitation to look up.

⚡ Demo & Code

🌐 Live Website (GitHub Pages): https://akashgt.github.io/wildquest-ai/
📱 Progressive Web App: Fully installable on iOS, Android, and macOS/Windows with offline caching.

akashgt / wildquest-ai

WildQuest AI — Explore more. Scroll less. An offline-first, mobile-friendly nature adventure app powered by open-weight AI for Hacktoberfest 2026.

WildQuest AI 🌿

Explore more. Scroll less.
A mobile-first, offline-first AI nature exploration application powered by Google’s open-weight Gemma model. Built for Hacktoberfest 2026’s “Touch Grass” open-source AI challenge.

📖 Table of Contents

🌿 Product Vision

Most modern digital products are designed to maximize screen time, engagement metrics, and endless scrolling. WildQuest AI does the exact opposite.

It is an offline-ready nature companion that invites you to leave your desk, step into your backyard, neighborhood park, or forest trail, and pay deep sensory attention to the natural world around you.

Instead of keeping your eyes fixed on a screen, WildQuest AI presents a single, mindful quest at a time with large, readable instructions…

View on GitHub

🏗️ How I Built It

WildQuest AI operates completely at the edge. The system is architected around a dual-engine pattern so that nature exploration is never interrupted by lack of cell service:

flowchart TD
    User([🌲 User in Nature]) --> Context[Select Environment & Duration]
    Context --> Orchestrator{AI Quest Coordinator}

    Orchestrator -->|Local Gemma via Ollama| Gemma[Google Gemma 2B]
    Orchestrator -->|No Ollama / Deep Woods| Fallback[Procedural Naturalist Engine]

    Gemma --> Zod[Zod Sanitizer & Safety Guardrails]
    Fallback --> Zod

    Zod --> UI[Adventure Mode]
    UI --> Journal[(IndexedDB Private Storage)]

1. Edge Intelligence via Google Gemma 2B

Connects directly to Gemma 2B running on your device via Ollama (http://localhost:11434/api/generate).
Strict JSON instruction prompts return structured micro-quests parsed through Zod schemas that strip unescaped HTML or unsafe markdown.

2. Dual-Engine Resilience (Offline Naturalist Engine)

The best trails have zero cell service. If local Ollama is offline or unreachable, WildQuest AI seamlessly routes to its built-in Naturalist Fallback Engine.
The app never fakes connectivity: the navbar status badge honestly reports whether Gemma or the Fallback engine generated your quest.

3. Non-Negotiable Safety Guardrails

🚫 Zero Foraging / Ingestion: Quests strictly prohibit tasting, eating, or picking wild plants or mushrooms.
🚫 Zero Wildlife Harassment: Always mandates safe viewing distances.
🚫 No Trespassing: Enforces public path awareness.
🚦 Street Safety: Instructs users to pause and step aside before looking up around traffic.

💡 Why Does Open Innovation Matter?

🌲 Trailheads Don't Have Cell Towers:

Cloud APIs from closed vendors break the second you hike into a canyon or state park. Open-weight models like Google Gemma run on local silicon—bringing intelligence to the wilderness without Wi-Fi.
🔒 Personal Reflections Remain Sacred:

Your thoughts, mental health pauses, and nature journal entries shouldn't train commercial models or feed behavioral ads. Open-weight models and IndexedDB keep your private life on your device.
💸 No Tollbooths on Nature:

Nature wellness shouldn't require an ongoing monthly API subscription or credit card. Open innovation democratizes tools so students, rangers, and community groups can run them freely.

🏆 Prize Categories

🌟 Featured Category: Best Use of Gemma ($200)

Local Edge Runtime: Tailored specifically for Google's open-weight Gemma 2B model on Ollama (gemma2:2b).
Deterministic Guardrails: Strict Zod schema validation ensuring safe, structured quest generation.
Truthful UI: Real-time Ollama status testing with explicit Gemma vs. fallback labeling.

🛠️ Partner Category: Best Use of GitHub ($100)

Automated CI/CD: Powered by deploy.yml running 17 Vitest unit tests, type-checking, and deploying PWA assets to GitHub Pages on every push.
Full Open-Source Compliance: Production Apache 2.0 license, clean issue templates, and comprehensive documentation.

🤖 My Agent Session

Built pair-programming with the Google Antigravity environment, using automated shell orchestration, test-driven validation, and browser-driven visual regression tests.

Explore more. Scroll less. Step outside and see what's blooming. 🌲

DEV Community: Akash Guha Thakurata