This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass.
What I Built
Pocket Field Notes turns four small choices—available time, setting, pace, and what you want to notice—into a printable three-step outdoor observation card. The point is to make the screen the shortest part of the experience. Read or print the card, choose a safe route yourself, put the device away, and look around.
I built it for people who want a low-pressure reason to step outside without tracking their location or following an app's route. The 10-, 20-, and 30-minute options work for a short walk or for sitting and observing from a doorstep.
Demo
Watch the 20-second demo. The field card in the video came from a real local run and its JSON response is included. The video uses rendered slides to show the flow; it is not a screen recording.
That run used 20 minutes · park · gentle walk · color. The card allotted 5 minutes to walk on a chosen route, 10 minutes to pause and observe, then 5 minutes to continue or return. Its reflection asks what colors and sounds stood out.
Code
Source, setup instructions, and MIT license. Install Ollama and the qwen2.5-coder:7b-instruct-q4_K_M model, run python server.py, then open http://127.0.0.1:8765. Python 3.11+ is the only app runtime dependency.
How I Built It
The browser sends four choices to a tiny Python standard-library server. The server calls Ollama on localhost, asking an open-weight Qwen2.5-Coder model for JSON with a title, three observation questions, and a reflection. The app checks the response shape. Deterministic code sets the step timing and movement, and replaces questions that assume a landmark the app has never verified. The browser renders and prints the card.
I tested the live app with the installed local model and adjusted the guardrails after an early output assumed a park entrance and pond. The final example avoids those claims. Codex assisted with the new code and this write-up; I reviewed and tested the running result.
Why Does Open Innovation Matter?
A local open-weight model can tailor an observation card without sending a user's location or preferences to a hosted inference API. The model is replaceable via OLLAMA_MODEL; the timing and safety rules remain visible in the repository. Once the model is installed, inference can run without an internet connection. There is no account, API key, usage charge, GPS, or tracking.
The model cannot know current conditions, opening hours, or whether a path is safe. The user chooses the place and route. That limitation shaped the product: it suggests what to notice, then gets out of the way.
Top comments (0)