This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass Gemma TrailMate 🌿 is a voice-first outdoor AI companion built around one absolute principle: “The screen should be the shortest part of the experience.”
Existing AI assistants often pull us deeper into our screens. Gemma TrailMate does the opposite: it uses open-source AI intelligence to synthesize highly customized, short, and active outdoor missions, then collapses its entire interface into a minimalist countdown timer and checkbox sheet. It encourages users to lock their phones, slip them into their pockets, step out, experience nature, and return only when finished to record their experiences.
[https://ais-pre-kdh7wmdrjmfra363y6dee5-718654813775.asia-southeast1.run.app](https://ais-pre-kdh7wmdrjmfra363y6dee5-718654813775.asia-southeast1.run.app)
[https://github.com/krithikgokuls/gemma-trailmate](https://github.com/krithikgokuls/gemma-trailmate)
AI Layer: Google's Gemma open-weight model family. We created custom system prompts that mold the LLM specifically to Gemma's 9B IT persona and concise, active-voice tone.
Frontend: React, TypeScript, and Tailwind CSS v4 utilizing an earthy, minimal visual design.
Backend: Express.js proxy routes mounting Vite dev middleware to allow unified full-stack serving on a single port.
Offline Resilience: Offline caching logic automatically loads preset local missions and nature guides during trail connectivity drops.
Privacy Controls: Zero external tracking databases are used. Private journals and photos remain strictly inside the user's browser local storage.
Getting people outside should not require sending every personal reflection, observation, or location coordinate to a closed, multi-billion-parameter cloud monopoly.
Open-weight models like Gemma put developers and users back in control:
Airgapped Privacy: Gemma's compact size allows it to run entirely locally (via Ollama or NPUs) in your browser or phone, guaranteeing your personal thoughts never leave your physical device.
Trailhead Portability: It enables applications to continue running fully offline in wilderness areas with zero network connectivity by hosting the weights on local trail routers.
Customizability: Developers can fine-tune Gemma specifically on regional botanical datasets, creating localized forest, desert, or mountain guides without being bound to generic closed-API rules.
You can view the full agent development session and iterative build history here:
Google AI Studio Developer Session: cb86a0bf-3ab4-470b-93d9-24825a466628 Grand Prize Category: Touch Grass AI / Open Source Innovation (Primary)
Gemma / Open-Weight AI Developer Track