Inspiration
Every week, teachers who collect work on paper face the same chore before marking even starts: open each photo or scan, work out whose it is, tick the roster, chase the missing students and rename files. Student work and names are personal data, so uploading them to a cloud grader is not an easy choice for a school. A NAS that already stores the files and runs AI locally is the natural place for this work.
What it does
Hand-in Desk is a teacher-operated Skill for UGREEN AI NAS. For one class and one assignment it:
- reads the header of each photographed or scanned page (name, class, roster number such as No. Absen, NIS or NIM, page number) with the OCR and vision capabilities of the NAS;
- matches pages to the class roster with rules for Indonesian and English classrooms, and reports who handed in, who is missing, duplicates, page gaps and pages that need the teacher;
- after the teacher approves an exact plan, moves pages into an assignment folder under self-describing names, or under their original names if the teacher prefers;
- optionally, right away or later, drafts rubric-based, evidence-quoted feedback and a draft grade table in Indonesian or English, for the teacher to edit and approve;
- saves a versioned hand-in record with a ledger of SHA-256 fingerprints, so the next run reads only new or late work, even when pages from earlier assignments are still in the inbox. It never deletes or overwrites files and never identifies anyone by face. It has no step that sends data outside the NAS, and its instructions tell the NAS experts to use on-device models only.
How we built it
With the UGOS Skill Creation Tool (uliya-skill-kit 0.7.3). Every step was checked against the public capability catalog, the Skill delegates only to registered experts (file-agent, document-agent, photo-agent, knowledge-agent), and the package passes the kit's static checker with no errors or warnings. The package is Markdown only: SKILL.md plus nine references. Claude Code (Anthropic) was used as an AI assistant for drafting and checking.
Challenges
Real classroom input: rotated or blurred phone photos, multi-page work, scanned PDFs that hold a whole class, Indonesian names with abbreviations and old spellings, one shared inbox for many assignments, and large classes, while keeping every write behind teacher approval and making re-runs safe.
What's next
If selected as a finalist: validate on the iDX6011 Pro with synthetic worksheets (two core tests and one edge test), measure the UGOS approval prompts, tune batch sizes and record the demo.
Built With
- markdown
- ugos
- ugreen-ai-nas
- uliya-skill-kit
