This is a submission for the Hacktoberfest Open-Source AI Challenge: Week 1 — Touch Grass.
What I Built
TrailLens Audio is a mobile-first trail companion designed to keep people outdoors instead of staring at a screen.
A hiker takes one photo of a plant, insect, fungus, or animal. TrailLens creates a careful field note, narrates a short naturalist guide, and then offers a Pocket Quest: a 20-second observation prompt that asks the hiker to put their phone away and reconnect with the trail.
Demo
Watch the outdoor TrailLens Audio demo
In the demo, I photographed a real trail specimen on my phone, received a local identification and audio guide, then started a Pocket Quest and put the phone away.
Code
View TrailLens Audio on GitHub
How I Built It
TrailLens Audio uses:
- Gemma 3 4B, running locally through Ollama, to inspect trail photos and create structured field observations.
- Kokoro 82M, a local neural text-to-speech model, for calm narration without a paid voice service.
- FastAPI and a lightweight mobile web interface for camera capture, playback, and Pocket Quest mode.
- A confidence-and-evidence layer that avoids presenting uncertain guesses as confirmed species identifications.
The app is intentionally cautious. If a photo is blurry, partial, or ambiguous, TrailLens labels it as unconfirmed instead of inventing a confident species name. It never recommends touching, eating, or disturbing a specimen.
Why Open Innovation Matters
Trails, gardens, and nature reserves are exactly where cloud-only AI becomes least reliable. Connectivity disappears, mobile data can be expensive, and uploading sensitive ecological locations is not always appropriate.
Running Gemma 3 locally makes TrailLens useful without internet access. Photos stay on local hardware, there is no per-photo inference cost, and the model can be swapped or improved later without rebuilding the app.
Open models made the central idea possible: a private, offline-friendly field companion that helps people look up from their phones instead of keeping them on-screen.
The Touch Grass Moment
The feature I am most proud of is Pocket Quest.
After each finding, TrailLens creates a short audio-only prompt that invites the user to notice leaf shapes, textures, sounds, or signs of shelter nearby. The screen fades into an “Eyes up. Trail on.” view.
TrailLens is not trying to make nature into another feed. It is designed to make the phone disappear as quickly as possible.
Prize Categories
- Best Use of Gemma
Top comments (0)