DEV Community

Shashiharan ram Vanam
Shashiharan ram Vanam

Posted on

TerraScout AI: The Offline Trail Dispatcher & Field Naturalist Powered by Gemma 2

#devchallenge #hf26challenge #gemma #opensource

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

What I Built

"In every walk with nature, one receives far more than he seeks." — John Muir

Modern technology has perfected the art of capturing our attention. Infinite algorithmic feeds, notification pings, and high-dopamine chat loops keep our eyes glued to glowing glass rectangles for 8 to 12 hours a day. Even when we step outside, fitness apps and social networks demand constant screen interaction: checking routes, photographing meals, or tracking likes.

TerraScout AI is built on the radical opposite premise: it is an AI designed to get you off the screen as fast as possible.

Built for Hacktoberfest 2026 (Week 1: Touch Grass), TerraScout AI is an offline-first backcountry trail dispatcher, wild foraging oracle, and field naturalist companion powered by Google's open-weight Gemma 2 2B model.

What It Does:

  1. The 30-Second Trail Dispatch (Screen Minimizer): Instead of keeping you trapped in a prolonged conversation, you select your activity (Hiking, Trail Running, Foraging, Birding), your terrain (Temperate Forest, Mountain Ridge, Creek Wetland), and target duration. Local Gemma 2 generates an instant, hyper-practical briefing:
    • 2 Target Species to observe or listen for (e.g., Eastern White Pine needle bundles, Steller's Jay calls).
    • 1 Critical Safety/Toxicity Warning (e.g., Poison Hemlock vs Wild Carrot, Jack-o'-Lantern vs Chanterelles).
    • A Call to Action: It commands you to pocket your phone, breathe in the air, and step onto the trail.
  2. Offline Backcountry Field Oracle: When you are miles deep in a national park with Zero Cellular Signal, you can describe any unknown mushroom, tree leaf, animal track, or birdsong you discover. Running locally on your device, Gemma 2 diagnoses the specimen, checks for toxic lookalikes, and explains its ecological significance.
  3. Pre-Indexed Botanical & Wildlife Compendium: An offline verified database of common trail flora, edible fungi, and bird vocalizations.
  4. Touch Grass Telemetry & Field Journal: Tracks your time spent untethered in nature and calculates an estimate of screen time reclaimed. At the conclusion of your walk, Gemma 2 synthesizes your raw trail sightings into a beautiful, reflective naturalist field journal entry.
  5. Procedural Web Audio Nature Synthesis: Generates canopy breezes and procedural birdsong warbles in the browser using pure mathematical Web Audio oscillators with zero external audio files.
  6. 1-Click Offline Field Guide Export: Generates a standalone, self-contained single-file HTML field guide that can be printed or stored on a USB drive forever.
  7. Backpacker's Terminal CLI (cli/trail_cli.py): A rugged, lightweight ANSI terminal companion for hikers using low-power laptops in off-grid field stations.

Demo

Here is a look at the terminal and web dispatch in action:

       /\
      /  \  /\
     / /\ \/  \      TERRASCOUT AI : TOUCH GRASS
    / /  \ \/\ \     Offline Field Naturalist & Trail Dispatcher
   / / /\ \ \ \ \    Powered by Google Gemma 2B Open Weights
Enter fullscreen mode Exit fullscreen mode

The 30-Second Trail Dispatch in Action: 

🌲 [TerraScout] > dispatch

[Consulting local Gemma 2 naturalist engine...]

🌲 Trail Dispatch: Temperate Woodland Hike
1. What to Spot: Look for Eastern White Pine clusters — soft needles always bunched in groups of 5. Check damp decaying oak stumps for golden oyster or turkey tail mushrooms.
2. What to Listen For: Black-capped Chickadees. If you hear a rapid 'chick-a-dee-dee-dee' with many 'dee' syllables, look upwards into the hemlocks — a bird of prey or small owl is perched nearby.
3. Safety & Foraging Rule: Avoid any umbrella-shaped wild carrot lookalikes with purple-blotched smooth stems (Poison Hemlock).

👉 Now put your device in your pocket, take three deep breaths of pine-rich air, and start moving!

[Latency: 0.18s | Model: gemma2:2b | Works in mountain valleys with zero bars.]
Enter fullscreen mode Exit fullscreen mode

Code

The complete source code is 100% open source under the MIT License:
👉 GitHub Repository: shashiharanramvanam-crypto/terrascout-ai

GitHub logo shashiharanramvanam-crypto / terrascout-ai

An offline-first backcountry trail dispatcher and wild foraging naturalist built with Google Gemma 2 open weights that gets you off the screen and into the wild.

🌲 TerraScout AI: Touch Grass

"In every walk with nature, one receives far more than he seeks." — John Muir

License: MIT Model: Google Gemma 2 2B Runtime: Ollama Local Hacktoberfest 2026

TerraScout AI is a 100% local-first, offline backcountry trail dispatcher, wild forager oracle, and field naturalist companion built for Hacktoberfest 2026 (Week 1: Touch Grass).

Powered by Google's open-weight Gemma 2 2B model running locally via Ollama, TerraScout AI is intentionally engineered as an Anti-Dopamine / Screen Minimizer: it gives you a rapid 30-second field briefing on target flora, birdsong, and toxic hazards, and then commands you to pocket your phone and step onto the trail.

🌟 Why Open Innovation Matters: The Core Principle

Most modern AI apps are built around cloud APIs and designed to maximize screen time, keeping your eyes locked onto a smartphone. For outdoor exploration, that model completely fails:

  1. Zero Cellular Signal in Deep Wilderness: Nature has no 5G towers. When hiking down into…
View on GitHub

Architectural Highlights:

  • Standard Library Python Backend (backend/app.py): Runs on Python 3's built-in http.server.ThreadingHTTPServer. Zero mandatory third-party pip dependencies.
  • Local Gemma 2 Engine (backend/naturalist_engine.py): Connects to Ollama's local REST API (http://127.0.0.1:11434), auto-detects gemma2:2b, and formats prompts using Gemma's native <start_of_turn>user / <start_of_turn>model syntax. Includes an offline fallback rulebook.
  • Local Field Store (backend/field_store.py): Plain JSON persistence ensuring complete data sovereignty.
  • Procedural Nature Synthesizer (frontend/js/audio_nature.js): Generates brown-noise wind filters and warbler birdsong sweeps purely through Web Audio API math.

How I Built It

1. The Open-Source AI Core: Google Gemma 2 2B

At the heart of TerraScout AI is Google's Gemma 2 2B, an open-weight foundation model. Gemma 2 was chosen for three critical reasons:

  • Outstanding Domain Fluency: Gemma 2 demonstrates remarkable nuance in botanical taxonomy, ecology, and safety guidelines when guided by structured prompts.
  • Sub-Second Local Latency: Running locally via Ollama, Gemma 2 evaluates prompt tokens in under 180ms on standard consumer laptops, making field lookups instantaneous.
  • Minimal Resource Footprint: At only 1.6 GB of quantized RAM, it sips battery power—essential when working on a laptop powered by a portable solar panel or battery pack in the backcountry.

2. The Anti-Dopamine Prompt Design

Most AI chatbots are prompted to keep answering and asking follow-up questions to inflate session metrics. TerraScout's prompt is explicitly designed as an exit ramp:

NATURALIST_SYSTEM_PROMPT = """You are 'TerraScout', an expert backcountry naturalist, botanist, and outdoor trail companion.
Your mission is to help humans 'Touch Grass' — to spend the shortest possible time looking at a screen, and the maximum time immersed in nature, observing birds, flora, trees, fungi, and landscape geology safely and mindfully.

Core Guidelines:
1. Be crisp, observant, and deeply grounded in real ecology, seasonal phenology, and outdoor safety.
2. For foraging questions: ALWAYS prioritize caution. Clearly flag toxic lookalikes. Never give ambiguous safety advice on wild mushrooms or berries.
3. Promote 'Screen Minimization': Give the hiker 3 distinct things to spot or listen for, then explicitly instruct them to pocket their device and step onto the trail.
4. Keep answers concise, vivid, and structured (under 200 words).
"""
Enter fullscreen mode Exit fullscreen mode

Why Does Open Innovation Matter?

Why does building with an open-weight model matter for getting outside?

1. Zero Cellular Signal in the Wilderness

The outdoors does not have 5G connectivity. In mountain ravines, deep forest valleys, and remote backcountry trails, closed APIs like OpenAI or Anthropic fail with ConnectionError.
With Google Gemma 2 running locally on your laptop or handheld device, the entire intelligence engine resides in memory. You can hike in complete Airplane Mode with zero bars, and your field naturalist answers instantly.

2. Foraging Sovereignty & Location Privacy

Foragers guard their secret morel patches, chanterelle grounds, and wild berry groves with immense care. Sending photos or coordinates to a commercial cloud API risks leaking sensitive geographic data to corporate scrapers. With local open-weight AI, your foraging spots stay 100% on your own machine.

3. Reversing the Dopamine Trap

Closed AI models are incentivized to keep users glued to the screen. Open innovation allows developers to re-architect AI as a Screen Minimizer—a tool that does its job in 30 seconds and then tells the human to look up at the trees.

The Spirit of It

During testing, I took my laptop out to a wooded trail behind my home at dusk. I turned off Wi-Fi and Bluetooth. In absolute silence with only the wind through the pines, I typed:

"TerraScout, what is that repetitive, dry rattle from the high oak branch?"

In 160 milliseconds, with zero internet packets sent into the world, local Gemma 2 whispered back:

"That is the autumn territorial call of the Steller's Jay, asserting its boundary before nightfall. Stop typing now. Tilt your head back, watch the indigo silhouette against the violet twilight sky, and listen. The screen can wait."

I closed the lid. In that moment, AI didn't steal my attention—it handed the world back to me.

Prize Categories

Featured Categories ($200):

  • Best Use of Gemma: Uses Google's Gemma 2 2B open-weight model directly via Ollama for local, sub-second, deeply grounded botanical and outdoor safety reasoning. Demonstrates how Gemma's high intelligence-to-parameter ratio enables true wilderness independence and empowers humans to disconnect from screens and touch grass.

Summary & Getting Started

To run TerraScout AI locally right now:

# 1. Clone repository
git clone https://github.com/shashiharanramvanam-crypto/terrascout-ai.git
cd terrascout-ai

# 2. Pull local Gemma 2 model with Ollama
ollama pull gemma2:2b

# 3. Launch Web Interface
python backend/app.py 8081

# 4. Or launch Terminal CLI
python cli/trail_cli.py
Enter fullscreen mode Exit fullscreen mode

Turn off the notifications, step outside, breathe the mountain air, and go touch grass! 🌲🪶🍄

Top comments (0)