DEV Community

Muhammadh Ahzem
Muhammadh Ahzem

Posted on

GemTrail: The AI That Tells You to Put Your Phone Away

#devchallenge #hf26challenge

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Submission 🌿

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

What I Built

Most modern AI applications are designed to optimize for screen time—infinite chat feeds, constant notifications, and endless scrolling.

GemTrail was built around the exact opposite philosophy:

"The AI should help the user explore the real world, not keep the user staring at the screen."

GemTrail is an AI-powered outdoor field observation companion. Instead of an unreliable computer-vision camera classifier that guesses gemstone names with false certainty, GemTrail serves as a patient natural observation coach:

  1. Screen Break / Field Mode: It presents a 5-minute outdoor field mission (e.g. Two-Tone Rock Explorer, The Tactile Hunter, or Sunlight Sparkle Search) and intentionally instructs the user to put their phone in their pocket and step outside.
  2. Guided Sensory Intake: When the user returns, they log what they directly felt and observed—color gradients, grain texture, angularity, hydrothermal quartz veins, or sunlight specular reflections.
  3. Local AI Observation Coach: Powered by Gemma 3 1B running completely on-device via Ollama, GemTrail explains geological mechanisms as possibilities (sedimentary strata, metamorphic foliation, hydraulic tumbling) without pretending to make definitive laboratory diagnoses.
  4. Actionable Next Outdoor Test: It prompts the user with one specific physical check to perform outside (e.g. tilting under sunlight to test cleavage planes, or wetting the rock to reveal hidden grain boundaries).
  5. Private Field Journal: Kept entirely in the browser's localStorage—no accounts, no tracking, and no cloud databases.

It is made for hikers, curious neighborhood explorers, students, and anyone who wants to rediscover the physical geology right under their feet.

Demo

Demo Highlights:

  • 0:00–0:20: Introduction & Philosophy — why AI should encourage outdoor presence.
  • 0:20–0:45: Field Mode Screen Break — starting a 5-minute outdoor timer with breathing animations.
  • 0:45–1:15: Stepping outside — finding a natural specimen and observing colors, bands, and textures.
  • 1:15–1:45: Recording observations & consulting Gemma 3 1B on-device.
  • 1:45–2:00: Terminal proof showing ollama list with gemma3:1b running privately with zero cloud API keys.

Code

GitHub logo Ahzem / gemtrial

🌿 GemTrail

An AI-powered outdoor field companion that helps people observe, investigate, and document rocks, minerals, and natural objects they discover outside.

Next.js React TypeScript Tailwind CSS Ollama License

đź’ˇ The Core Philosophy

Most AI applications are designed to keep you glued to a screen. GemTrail is designed around the opposite idea:

"The AI should help the user explore the real world, not keep the user staring at the screen."

Instead of building a computer-vision camera classifier that guesses gemstone names with false certainty, GemTrail acts as an outdoor observation coach:

  • It prompts you to put your phone away.
  • It guides you on 5-minute outdoor missions.
  • It teaches you how to look for cleavage planes, hydrothermal veins, striations, and sunlight glints with your own eyes.
  • It runs an open-weight model (Gemma 3 1B) 100% locally on your machine—ensuring complete privacy and zero cloud dependencies.

✨ Features

  • ⏱️ Field Mode & Screen Break: A…
View on GitHub

Repository link: https://github.com/Ahzem/gemtrial

The complete stack is open-source:

  • Next.js 16 (App Router) + React 19 + TypeScript
  • Tailwind CSS v4 with glassmorphism and outdoor dark-mode aesthetics
  • Ollama API Client calling gemma3:1b with strict structured JSON output and fallback heuristics
  • Web Audio API synthesized chimes for the Field Mode timer

How I Built It

1. Local Open-Weight Model: Gemma 3 1B on Ollama

I chose Gemma 3 1B, Google's lightweight open-weight model, served via the local Ollama daemon (http://127.0.0.1:11434). At only ~815 MB, it loads fast on ordinary consumer hardware and delivers near-instant inference with zero cloud latency and zero API fees.

2. Guardrailed Observation Coach Prompt

Rather than asking the model to hallucinate mineral names, I engineered a disciplined system prompt:

You are GemTrail, an outdoor field observation companion.
You are an observation coach, NOT a professional geologist or gemologist.

IMPORTANT RULES:
- Never claim that you can positively identify a rock, mineral, or gemstone from a simple description.
- Never invent scientific facts about an unknown object.
- Clearly distinguish observations from possibilities.
- Encourage the user to observe the object themselves outside.
- Give the user one useful next observation to make in the sunlight.
- Remind users to practice Leave No Trace.
Enter fullscreen mode Exit fullscreen mode

The route handler (app/api/gemtrail/route.ts) validates connection health, queries the local model, parses structured responses, and logs latency.

3. "Touch Grass" Screen Break UX

The application is built around the FieldModeScreenBreak component. When a user accepts a mission, the app transitions into an ambient field display, initiates a 5-minute timer with a gentle breathing pulse, and prompts:

"Put your phone away. Spend 5 minutes observing. We'll be here when you return."

Why Does Open Innovation Matter?

Open innovation is the entire backbone of what makes GemTrail viable and ethical:

  1. True Privacy in Nature: Closed APIs require transmitting personal exploration notes, photos, and timestamps to remote corporate servers. By using open-weight Gemma 3 running locally on Ollama, observations never leave the user's machine.
  2. Offline & Remote Capability: Nature trails and state parks rarely have dependable 5G connectivity. Open-weight models empower users to bring intelligent observation tools into off-grid wilderness environments.
  3. No Financial Gatekeeping: Closed LLMs require recurring API keys and credit cards. Open models let students, park rangers, and backyard hobbyists explore science freely without fearing a cloud billing spike.
  4. Resilience & Independence: When an app is built around open weights and local code, it never breaks because an upstream API changed its pricing tier or discontinued an endpoint.

My Agent Session

Developed with pair-programming assistance from Antigravity IDE, testing TypeScript schemas, Next.js 16 routing conventions, and live local Ollama model verification.

Prize Categories

  • Touch Grass Track (Getting developers and users outdoors into nature)
  • Open Innovation Track (Harnessing open-weight Gemma 3 1B for local, private, on-device AI)

Top comments (0)