This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
TrailLens is an offline plant identifier. You take a photo of a plant
and a local AI model tells you what it likely is, with key features and
a safety note, with no signal or internet needed. It's for hikers,
gardeners and anyone curious about what's growing around them. The
screen part takes seconds, then you keep walking.
Demo
traillens.netlify.app
Code
https://github.com/somyamoonat/hactoberfest-week-1
How I Built It
- Gemma 3 (open-weight vision model) for identification
- Ollama for local inference on my MacBook Air M5
- Streamlit for the interface
- Prompted the model to return structured JSON, and saved results to a local field journal
Why Does Open Innovation Matter?
- It works with wifi off, which matters on a trail with no signal.
- My photos never leave my laptop.
- It costs nothing per scan.
- I can swap the model by changing one line.
My Agent Session
I built TrailLens with Antigravity as my AI coding agent, running on a MacBook Air M5.
What I asked it to build:
A Streamlit app that identifies plants from a photo using a local vision model (Gemma 3 via Ollama), with no cloud APIs. It returns structured JSON (name, features, confidence, safety warning) and saves results to a local field journal.
How the session went:
- I gave Antigravity one detailed spec covering the stack, features, and safety banner.
- It generated the app, README, MIT license, and requirements.
- I ran it, tested it with wifi off, and fixed issues as they came up.
- I then took it outside and tested it on real plants. The agent wrote the code, but the open-source pieces (Gemma 3 and Ollama) are what make the app work offline.
Team member: Nilay raj
dev username: nilay_raj
Top comments (0)