Field-Note: Turning Real-World Observations into Evidence-Backed AI Drafts

title: "Field-Note: Turning Real-World Observations into Evidence-Backed AI Drafts"
published: false
tags: devchallenge, hf26challenge

This is a submission for the "Hacktoberfest Open-Source AI Challenge — Week 1: Touch Grass" (https://dev.to/challenges/hacktoberfest-week1-2026-10-05).

What I Built

Field-Note is a mobile-friendly field observation application designed to help people record real-world activities, preserve supporting evidence, and turn their observations into structured, AI-assisted draft findings.

Fieldwork often involves taking photographs, recording notes, and documenting activities on construction sites or during inspections. Keeping these details organized and connecting observations to their supporting evidence can be difficult.

Field-Note brings these elements together in one place.

Users can:

• Capture field notes with a title and written description.
• Attach photographic and audio evidence.
• Optionally record their location.
• Save observations locally in their browser.
• Request AI-assisted analysis to generate structured draft findings when the backend and AI provider are configured.

The central principle behind Field-Note is simple: AI proposes. Evidence proves. Humans verify.

AI-generated findings remain drafts rather than verified facts. The original observation and its evidence remain the basis for human review.

Demo

Live application: "Open Field-Note" (https://special-potato-x55797pvqrrrf6q5x-5173.app.github.dev/observations/af775afd-7b78-4b18-a266-bb4be0f0eccf)

Note: This is a temporary GitHub Codespaces development URL. The application may be unavailable when the Codespace is stopped. The linked observation is an example from the development environment, not a permanent public demo.

Demo video: To be added if available.

Code

GitHub repository: https://github.com/Shorunkemanuel/Field-Note

Field-Note is developed as an open-source project. The repository contains the application source code, backend API, frontend, and tests.

How I Built It

Field-Note uses a React and TypeScript frontend, Vite for development and building, and a Python backend built with FastAPI.

The application uses browser-based storage to preserve field observations and their evidence on the user's device. The backend provides APIs for retrieving observations and requesting AI-assisted extraction.

The AI integration uses an OpenAI-compatible provider interface. This allows the backend to communicate with compatible inference services through server-side configuration rather than exposing API credentials in the frontend.

The extraction workflow is designed to produce structured findings containing descriptions, categories, supporting evidence references, reasoning, uncertainty, confidence where available, and suggested actions.

The application also validates the model's structured output and checks that referenced evidence belongs to the material actually analyzed.

AI-generated findings are explicitly marked as drafts. They are not automatically treated as verified inspection results.

The project is still being tested, particularly the external AI provider integration. The core application and its AI extraction workflow should be described according to the functionality verified in the final demo.

Why Does Open Innovation Matter?

Field-Note explores how open-source software and open-weight AI models can support practical tools for people working outside traditional office environments.

Open innovation gives developers the flexibility to experiment with different model providers, inspect the application architecture, adapt the software to their needs, and build on existing technologies.

A provider-independent integration also makes it easier to evaluate alternative inference services instead of permanently tying the application to one vendor.

For fieldwork, the goal is not simply to generate text with AI. It is to make observations easier to organize while maintaining a clear connection between AI-generated suggestions and the evidence that supports them.

Keeping AI findings as drafts helps preserve human judgment and accountability.

My Agent Session

Agent session: Not included yet.

Prize Categories

Hacktoberfest Open-Source AI Challenge — Week 1: Touch Grass.

Additional categories should be listed only if the project meets their eligibility requirements.