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

What I Built

Most modern software is deliberately engineered to maximize your screen dwell time. The longer your retinas stay glued to pixels, the more profitable the app becomes.

TrailGemma does the exact opposite.

I built TrailGemma under a singular design constraint: The best outdoor AI build must make the screen the shortest part of the experience.

TrailGemma is an audio-first nature and outdoor trail companion designed for hikers, trail runners, birdwatchers, and anyone suffering from digital fatigue. You spend just 15 seconds interacting with the screen:

1. Select your outdoor vibe (Fall Foliage Drift, Morning Songbird Chorus, Shinrin-Yoku Forest Bath, or Trail Tempo Run).
2. Choose your walk duration (10 to 60 minutes).
3. Tap "Enter Pocket Mode" and slide your phone into your pocket.

From that moment on, your phone screen dims into a calm, ambient breathing orb. As you hike, run, or walk, TrailGemma's audio engine delivers:

• Procedural Ambient Forest Soundscapes: Synthesized canopy wind, rustling autumn leaves, and native bird chirps generated in real-time via the Web Audio API without needing external MP3 downloads.
• Timed Spoken Audio Checkpoints: Sensory nature guides that prompt you to look up at the tree canopy, smell the pine needle floor, and feel the mountain air.
• "Touch Grass" Micro-Quests: Tactile challenges at every checkpoint (e.g., "Touch the damp moss on the north side of the bark. Notice how much cooler it feels than the surrounding air.").
• TrailLens Specimen Scanner: If you spot an interesting leaf, bracket fungus, or bird track, Gemma identifies its taxonomy, foraging safety, ecological role, and observation tips with spoken audio.

Demo

• Live Application (Hosted on Render): https://trailgemma.onrender.com
• GitHub Repository: https://github.com/khushalkks/TrailGemma

Core User Experiences:

1. Pocket Mode (Screen-Off Guide): Locks and dims the screen so you never feel the urge to scroll or check social media outdoors.
2. TrailLens Specimen Identifier: Instant botanical and wildlife identification powered by Gemma with zero-signal edge knowledge.
3. Scenic Radar & Map: Interactive autumn foliage map with waypoints, nature soundscapes, and audio checkpoints.
4. Procedural Nature Soundscapes: Real-time Web Audio API sound synthesis (zero external media buffer failures).

Code

The entire codebase is 100% open-source under the MIT license:

How I Built It

TrailGemma was built using open-source AI and partner runtimes:

1. Open-Weight Core (Google Gemma 2):
   • We leveraged Google's open-weight Gemma 2 (gemma-2-9b-it) to structure dynamic sensory trail itineraries, safety advisories, and botanical specimen analysis.
   • For edge inference, the architecture supports local Ollama (gemma2:2b) as well as an edge-cached taxonomy database that functions with zero cellular signal.
2. Production Runtime & Hosting (Render):
   • The entire full-stack application and API are hosted live on Render as a web service at https://trailgemma.onrender.com.
   • Configured with a dedicated render.yaml blueprint and a lightweight multi-stage Docker container for effortless automated deployments and health checks.
3. Voice Narration (ElevenLabs + Web Speech API):
   • Natural woodland ranger voice narration via ElevenLabs TTS API (eleven_turbo_v2_5), backed by an automatic client-side Web SpeechSynthesis fallback so the guide functions even in remote wilderness.
4. Procedural Nature Soundscapes (Web Audio API):
   • Rather than relying on bulky external audio files that fail to load on weak trail connections, we created a procedural sound engine in pure Web Audio API that synthesizes pink-noise canopy wind, resonant gust filters, and realistic songbird chirps in real time.

Why Does Open Innovation Matter?

This challenge asks: Where does an open-based approach work better than a closed one?

For an outdoor companion, open innovation is not just a nice-to-have; it is the only viable architecture:

1. Closed Cloud APIs Fail in Deep Woods: When you hike into a national park, mountain hollow, or dense forest, there are 0 bars of cellular service. If your app relies on proprietary cloud APIs like ChatGPT or Claude, it immediately throws a connection timeout and crashes. Open-weight models like Gemma can run on edge devices, local laptops, or cached runtimes with zero internet.
2. True Location & GPS Privacy: Outdoor enthusiasts and foragers often visit sensitive ecological habitats or secret wild mushroom spots. Sending continuous GPS coordinates and trail voice queries to corporate cloud servers creates invasive location footprints. With open-source models, your trail data and coordinates never leave your device.
3. Zero Marginal Token Tax: A closed API meters every token you read or speak, costing recurring fees. Running open-weight Gemma locally or self-hosting on Render means community trail clubs, scouts, and hikers can use the companion unlimited times for $0.00.
4. Anti-Addictive by Design: Closed AI apps are incentivized to keep you engaged and looking at screens. With open-source code, we specifically engineered the app to shut off the screen and prompt you to touch physical grass, bark, and soil.

Prize Categories

I am submitting TrailGemma for the following categories:

• Best Use of Gemma ($200 USD): Google's open-weight Gemma model is the core intelligence behind the outdoor trail itineraries, ecological checkpoints, and botanical identification.
• Best Use of Render ($200 USD): The full-stack application and API are deployed and running live on Render (https://trailgemma.onrender.com) with a dedicated render.yaml blueprint, production Dockerfile, and health check monitoring.
• Best Use of ElevenLabs ($100 USD): Powers the realistic, calming outdoor ranger narration so hikers can keep their phones safely in their pockets.

Thanks to DEV, Hacktoberfest, Google Gemma, Render, and ElevenLabs for organizing this challenge! Put your phone away, step outside, and touch grass today!