RoyaCheck Offline

RoyaCheck Offline is a hackathon MVP for browser-local coffee-leaf observation. It lets a user select or capture a coffee-leaf image, run a compact local AI proposal on the device, review that proposal, choose a human disposition, and save a text-only local observation. The prototype is designed for low-connectivity field contexts where small AI should support human review without pretending to replace it.

• Live demo: https://royacheck-offline-current.onrender.com
• Public repository: https://github.com/SplitzHappen/RoyaCheck-Offline

What it does

RoyaCheck Offline accepts one image at a time and proposes one of three outputs:

• visible_rust
• no_visible_rust
• not_sure

The AI output is only a proposal. The user must review the image, choose a human disposition, confirm that choice, and save the local observation. Only the human disposition is treated as the formal local record.

How to use the live demo

1. Open the live demo link.
2. Select a coffee-leaf image from a computer, or on mobile choose an image from the gallery or take a new picture.
3. Run the local AI check.
4. Review the AI proposal and the image.
5. Choose the human disposition.
6. Confirm the human decision.
7. Save the text-only local observation.

Run locally

This is a static browser app. From the repository root:

python3 -m http.server 4173

Then open:

http://127.0.0.1:4173/app/

The repository also defines a convenience script:

npm run serve:app

Architecture summary

• Static browser application under app/.
• Local ONNX inference in the browser.
• ONNX Runtime Web is served with the app.
• No backend inference API is required.
• The browser-local workflow separates the AI proposal, the human disposition, and the saved local record.
• Local storage is used for saved text-only observations in the MVP flow.

Data and privacy note

The selected image is processed locally for inference. In this MVP, saved observations are text-only local records. The raw image is not saved with the record.

Human-final authority

RoyaCheck Offline is intentionally designed so that the model does not make the final decision. The model proposes; the human signs the record. The saved local observation is based on the human disposition, not automatic acceptance of the AI output.

Submission notices

The public submission notices are maintained under docs/submission/:

• AI_TOOLING_DISCLOSURE.md
• THIRD_PARTY_NOTICES.md
• SUPPORTED_BROWSER_AND_OFFLINE_LIMITATIONS.md

The canonical public deployment is the Render-hosted live demo linked above. GitHub Pages is not the canonical deployment surface for this submission.

Safety and claim boundaries

This prototype does not:

• diagnose a plant;
• recommend treatment;
• recommend pesticide use;
• claim field validation;
• claim WBG or World Bank Group endorsement;
• verify that the image is a coffee leaf;
• replace qualified human review.

The local-language component is a fixed-string scaffold only and remains pending fluent human validation.

Status

This is a hackathon prototype / MVP, not production software. Stage 7 and Stage 8 evidence records exist in the repository, but this README does not claim formal closure of those stages or add new validation, field-performance, or deployment-readiness claims.