1. One sentence: What does the Skill do?

It pairs undated family photo scans with recorded family stories on the NAS and builds a story album where every date, place and name cites its source.

2. Who is the target user and problem?

The family member who keeps the NAS archive. Scanned prints only carry the scan date, and only elderly relatives know the stories. Families record them, but the audio stays unheard and disconnected from the photos.

3. What is your proposed solution?

The Skill transcribes the recordings, reads each photo (date stamps, handwriting, scene, recognized people), links photos to story segments, and drafts a story card with a date range. The family approves each card; only then are cards saved and an album created. Originals stay untouched.

4. What is the Core User Flow?

  1. "Match grandma's recordings to the Grandma box scans."
  2. Skill confirms scope (214 photos, 6 recordings) and reads without writing.
  3. Cards arrive in batches: "Aug 1987, Lake Uiam. Sources: Interview-3 12:41 'right after the baby turned one'; stamp '87 8 14'."
  4. Gaps become questions: "When did you move to Busan?"
  5. User approves one write plan; cards and album appear.

5. What substantive difference does your idea offer compared with existing tools or conventional AI solutions?

File chat assistants answer a question and forget it, often without sources. This Skill leaves a lasting album and states nothing without a checkable source: a recording second, a print stamp, or the user's own answer. Dates are ranges with reasoning; "after the baby turned one" becomes a year only when another source anchors it. Weak matches become questions.

6. Which UGREEN AI NAS capabilities will be used, and at which step in the workflow?

Step Area Capability
Scope, hashes Files ug-filesystem find, stat, sha256sum; ug-filemgr mediainfo
Read photos Albums ug-photo search-picture, meta, vision-ocr, vision-vlm
Transcribe Knowledge ug-knowledge audio-transcription
Letters, diaries (optional) Documents ug-document extract
Save cards, manifest Files, Knowledge ug-filesystem mkdir; ug-knowledge save-text
Story album Albums ug-photo create-album
Replay quote (optional) Music ug-music search, control seek

7. What was your technical approach?

Built on the uliya-skill-kit 0.7.3 catalog. The Skill only delegates to five catalog experts, passing real paths, identities and segment times forward. Writes run only after card review and an approved write plan, under platform approval; the Skill has no delete, move, rename or overwrite step. Local NAS models do all AI work, so family voices never leave home. Scanner dates are discarded. All five delegation states are handled.

8. What is the current development status?

Implemented: SKILL.md with nine orchestrated steps, plus references for evidence rules, data formats, approvals and a walkthrough. It passes the kit's static checker with zero warnings. Awaiting NAS validation: transcription quality, model speed.

9. What is your final validation plan?

Test 1: 10 scans, 2 recordings, known answers; each approved fact cites the correct second and each date range contains the true date. Test 2: reject 3 cards and decline one write; nothing is written for them and original hashes match. Edge: a photo with only a scan date gets "no story found" and no write. Finalist demo: a scan, grandma's sentence, the evidence card, approval, the finished album.

10. Are there any known limitations of your Project?

Transcription may struggle with dialects and older voices. Local vision models are slow, so large boxes run in batches. Per-file write approvals could mean many confirmations. Replay works only for recordings in the Music library.

Built With

  • ai-nas
  • claude
  • ugos
  • ugreen
  • uliya-skill-kit
Share this project:

Updates

Submission history