This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
We have a paradox in the tech world: we build incredibly powerful AI that requires us to be tethered to our desks, bathed in the blue light of our monitors, and dependent on gigabit fiber connections.
For the "Touch Grass" challenge, I wanted to build the anti-app. An AI whose sole purpose is to make you look away from the screen, wander off the beaten path, and literally get your hands dirty.
Meet TrailSight.
What I Built
TrailSight is an offline-first, edge-computed AI foraging guide and trail naturalist. It is designed specifically for the deep woods, remote mountains, and hidden valleys where cell service goes to die.
You point your device at a strange mushroom, a set of animal tracks, or an unfamiliar berry. TrailSight analyzes the image, identifies the specimen, and provides critical context: Is this edible? Is it an invasive species? What bird makes that nest?
How it gets people off the screen:
TrailSight features a "Pocket Mode." Once you snap a photo, you put your phone away. The local AI processes the image and uses local Text-to-Speech (TTS) to whisper the findings into your earbuds while you continue your hike. The screen is just the camera shutter; the actual experience happens in the world around you.
It is built for hikers, foragers, amateur botanists, and anyone who wants to rediscover the wilderness without needing five bars of 5G.
Demo
Imagine a video playing here... 🎬
A dense, foggy trail in the Pacific Northwest. The phone screen clearly shows "No Service." The user snaps a picture of a vibrant orange mushroom on a rotting log, locks the phone, and puts it in their pocket. A few seconds later, an audio voiceover chimes in: "That appears to be a Chicken of the Woods. It is a choice edible mushroom. Notice the overlapping, sulfur-yellow shelves..." The user smiles, harvests a piece, and keeps walking.
Live Demo / Docs: TODO
Code
The entire stack is built to run locally on a smartphone or a Raspberry Pi field kit.
github 1Prototype1/TrailSight-TouchGrass
Core components include:
-
vision_engine.py: Handles local inference of the quantized vision model. -
audio_guide.py: The local TTS pipeline for screen-free feedback. -
trail_memory.sqlite: A local vector database that maps your discoveries to GPS coordinates (saved locally, synced only when you choose to).
Top comments (0)