Footpath Receipt — Small observations, clearly recorded

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

What I Built

Footpath Receipt turns details noticed on a familiar walk into clear, reusable field notes. I wanted a lightweight way to capture what I observed and organize it later, while keeping the person who took the walk in control of the result.

Write down what you personally noticed, and Gemma organizes those notes into an editable field receipt. Each observation points back to a note ID and an exact quote, with follow-up questions kept visible. You can edit, copy, or download the result.

The goal is to make the walk the important part and keep the screen task small: capture a few notes, organize them, then review the draft.

Demo

[https://drive.google.com/drive/folders/1vZmR8WIu4WR2xkfzcchvyYpjZRZQePoA?usp=sharing]

Code

Footpath Receipt on GitHub

How I Built It

The app uses plain HTML, CSS, browser JavaScript, and Node.js’s built-in HTTP server, so it starts without a framework or package installation. The server sends the notes and optional manually typed area label to Google’s hosted Gemini API, using the open-weight Gemma 4 26B A4B Instruct model (gemma-4-26b-a4b-it). The API key stays on the local server and isn’t included in the repository or browser code.

To keep results grounded in the author’s notes, the server checks each note ID and verifies that its quoted source text appears in the cited note. The interface shows that quote beside each organized observation, making it easy to review and refine the result.

The automated test suite passes four tests covering note validation, citation mapping, rejection of invented sources, and malformed model output. I also ran a real Gemma request on my Windows laptop and received a response, confirming the hosted inference flow end to end.

Why Does Open Innovation Matter?

Gemma’s open weights give developers the option to adapt or self-host the model under its terms. I chose hosted Gemma for this version to keep setup simple—there’s no model download. In hosted mode, the app sends the notes and optional area label to Google for generation, while keeping the API key on the local server. This gives the project a quick way to try open-weight AI today and a path to local deployment or adaptation later. The author reviews the cited draft before deciding how to use it.

Prize Categories

Best Use of Gemma — Gemma powers the project’s central task: organizing first-hand notes into a cited, editable field receipt. I verified the hosted inference flow with a real request on my Windows laptop.