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

What I Built

Trailtale is a local, offline walking app that turns a walk into an immersive audio story. While you move, the app plays a short story segment; when you stop, the audio pauses automatically. It also surfaces weekly nature quests and lets you confirm interesting finds with a photo, all without pulling you back to a screen.

It’s built for people who want to spend more time outside without getting stuck in their phone. The idea is simple: make the walk the main experience, not a distraction from it.

Code

Repository:

GitHub logo HarshitAnand1 / trailtale-touch-grass

Trailtale

An audio story that plays only while you walk, plus weekly nature quests. Every model runs locally — no cloud, no account, works offline.

Quick Start

1. Install Python dependencies

cd server pip install -r requirements.txt

2. Start the server

# From the project root uvicorn server.main:app --host 0.0.0.0 --port 8000 --reload

3. Open the app

Visit http://localhost:8000 on your phone (or laptop browser).

4. (Optional) Set up llama.cpp for story generation

# Download a Gemma model and start the llama.cpp server ./llama-server -m gemma-2b.gguf --port 8080

Without llama.cpp, the story engine returns fallback text so you can still test the full walk flow.

5. (Optional) Install Piper for TTS

# See https://github.com/rhasspy/piper for installation piper --model en_US-lessac-medium --output_file test.wav <<< "Hello from the trail"

Without Piper, the TTS engine generates silent placeholder audio.

Architecture

Phone (PWA) Host (FastAPI) ┌──────────────┐ ┌──────────────────┐ │ Walk Gate │──────▶ │ Orchestrator
…
View on GitHub



Project source: this includes the PWA client, FastAPI backend, quest data, and local AI pipeline.

How I Built It

I built Trailtale as a combination of a mobile-first PWA and a local AI backend:

  • Frontend: a lightweight browser app in the client/ folder using HTML, CSS, and JavaScript
  • Sensors: phone motion and GPS data are used to detect whether the user is walking or stopped
  • Backend: FastAPI in the server/ folder orchestrates the walk flow, story queue, quests, and logs
  • Story generation: a local model runs through llama.cpp using Gemma
  • Voice: Piper generates spoken audio from the generated story text
  • Quest system: weekly nature quests are stored as local JSON packs and tracked in SQLite
  • Optional vision check: a local multimodal model can verify that a photo matches a quest item
  • Offline-first design: the app is intended to work with no cloud service, no account, and no internet dependency

This stack makes the app feel like a real walk companion rather than a web app that happens to use AI.

Why Does Open Innovation Matter?

Open innovation mattered here because the whole product depends on privacy, local control, and flexibility.

A closed API would have forced:

  • cloud-based story generation for every segment
  • network calls for TTS and photo verification
  • higher cost per walk
  • user data leaving the device

With open-source tools and open-weight models, Trailtale can run locally on a phone, laptop, or a backpack setup. That means:

  • no cloud dependency
  • no account required
  • no pay-per-token barrier
  • the prompt, model, and voice can all be tuned or swapped easily

For an app designed to get people to step away from screens, local open tech is not just a technical choice—it’s the product requirement.

Prize Categories

Gemma