TouchGrass AI: Plan less, explore more.

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

What I Built

What TouchGrass AI Does

TouchGrass AI is a privacy-first, offline-first outdoor adventure companion designed to minimize screen time while helping you explore the physical world. It guides your walks using deterministic mapping and local open-weight AI rather than keeping you hooked on an addictive interface.

The end-to-end adventure cycle includes:

• Rapid Adventure Planning: Tailors a walk in 30 seconds based on your schedule, fitness level, and interests, generating custom sensory missions and safety checklists.
• Reliable Offline Routing: Caches verified route geometry and quests directly into device storage via OpenStreetMap and OSRM with zero hallucinations or fake coordinates.
• Minimalist Outdoor Mode: Transforms your interface into a pocket-first screen showing only a walk timer and active missions so you can put your phone away.
• Field Observations: Lets you snap photos using a field viewfinder with a rule-of-thirds grid and record voice notes via in-browser speech recognition.
• AI Digital Scrapbook: Synthesizes your route, completed missions, photos, and voice notes into a vintage journal complete with a field narrative and 1-click Markdown export.

Does It Get People Off the Screen and Into the World?

Yes, fundamentally. Its core design philosophy revolves around making the screen the shortest part of the experience. Instead of functioning as an endless social feed or SaaS dashboard, it uses a "pocket-first" approach:

• It actively encourages you to put your phone in your pocket during the walk via a distraction-free Outdoor Mode.
• It replaces mindless scrolling with tactile engagement through active sensory prompts (e.g., listening for birds or finding specific textures).
• It shifts the digital interaction to after the walk, using local AI to turn your real-world offline experiences into a digital scrapbook journal.

Who Is It For?

• Digital Minimalists & Tech Workers: Individuals looking for a intentional, low-distraction tool to unplug and step away from notification noise.
• Nature Enthusiasts & Explorers: People who want structured walking loops, sensory quests, and trail guides without relying on cloud-tracking fitness apps.
• Privacy Advocates: Users who want 100% local data control—ensuring location traces, photos, and voice notes never leave their device thanks to zero-cloud telemetry and edge AI models like Gemma 2 and Whisper.

Code

You can explore the complete codebase, backend logic, frontend architecture, and setup instructions directly via the repository here:

TouchGrassAI Repository on GitHub

How I Built It

TouchGrass AI combines lightweight open-weight models, deterministic routing infrastructure, and a local-first software architecture to balance offline autonomy with zero-cloud telemetry.

🤖 Open-Source AI & Models

• Text & Reasoning Engine: Powered by Google DeepMind Gemma 2 (2B) via Ollama, handling dynamic workout parameterization, personalized sensory quest generation, and safety checklist formatting.
• Vision & Field Identification: Uses Moondream (an edge vision-language model) running locally to process field photos and classify nature, flora, and wildlife observations.
• Voice & Transcription: Integrated with speech-to-text processing for recording field audio notes completely hands-free on-device.
• Zero-Dependency Fallback: Built with a deterministic rule-based fallback engine so the app remains fully functional even if local AI services are turned off or unavailable.

🗺️ Infrastructure & Routing

• Deterministic Geometry: Relies strictly on OpenStreetMap (OSM) and Project OSRM for routing data. This ensures all generated tracks, loops, and trails use physical pathways without hallucinations or fake coordinates.

⚙️ Software Architecture & Frameworks

• Backend: Built using FastAPI (Python) to manage API endpoints, route geometry requests, and local database handling.
• Frontend: A React + TypeScript + Vite progressive web app (PWA) structured for high performance and responsiveness.
• Offline Storage: Utilizes browser-based IndexedDB alongside a Service Worker to cache complete adventure packets (routes, quests, and observation schemas) for total offline usage and Airplane Mode compatibility.
• Data Storage: Uses a local SQLite database (touchgrass.db) to preserve a strict privacy boundary with zero cloud tracking or telemetry.

Why Does Open Innovation Matter?

Open innovation is essential for building technology that respects human attention and data sovereignty. It champions transparency, community-driven progress, and the fundamental right to own your stack without being locked into proprietary ecosystems.

For TouchGrass AI, open innovation isn't just a development preference—it is an architectural necessity. Here is why it matters and what it makes possible:

🚀 What Open Innovation Made Possible

• 100% Offline Capability & Airplane Mode Reliability
• Closed APIs: Require persistent cloud connectivity, streaming tokens back and forth over cellular networks. If you wander into a dead zone, a park, or a mountain trail, a closed cloud app breaks down entirely.

• Open Innovation: Running open-weight models locally via Ollama means the entire inference and routing engine lives on the device. You can turn on Airplane Mode, walk deep into nature, and still have fully functional AI quest generation and routing.

• Absolute Zero-Telemetry Privacy

• Closed APIs: Inevitably ingest user telemetry, location traces, photos, and voice memos to train corporate models. For an app designed around personal wellness and outdoor exploration, piping your physical location and journal entries to a third-party server defeats the purpose of unplugging.

• Open Innovation: Keeps all personal data strictly local inside SQLite and IndexedDB. Your walks, photos, and voice notes never leave your hardware.

• Deterministic Physical Grounding (Zero Hallucinations)

• Closed APIs: Proprietary mapping or generalized LLMs are prone to hallucinating paths, recommending impassable routes, or inventing fake coordinates.

• Open Innovation: Leveraging OpenStreetMap and OSRM guarantees deterministic, community-verified geographical geometry. You get real pathways, physical loops, and trails you can actually walk safely.

• Full Hackability and Extensibility

• Open sourcing the repository under an MIT License means anyone can inspect the codebase, audit the privacy safeguards, add custom local models, or tailor new sensory quest packs for their local geography.

Prize Categories

TouchGrass AI aligns with several impact and technology partner categories, specifically designed around edge-deployed open-weight models and offline-first community resilience:

• Global Resilience (Impact Track): Focuses on building offline, edge-based systems and resilient tools for outdoor or low-connectivity environments.
• Ollama (Special Technology Prize): Showcases local open-weight execution using Ollama to drive edge inference (utilizing Gemma 2 and Moondream).
• Local / On-Device Intelligence & Offline-First: Highlights architectures that eliminate cloud dependencies to ensure complete user privacy and field reliability in remote areas.