Event Memory Rebuilder

Inspiration

Family memories end up scattered. The photos are in one folder, the videos in another, and the invitation or ticket for the same day sits somewhere else. Putting one occasion back together, such as a festival at grandma's house or a trip, means searching all three by hand, so most people never do it.

A NAS already keeps all of this privately at home. That made me ask: could an AI Skill rebuild a whole event for you without any of it leaving the device?

What it does

You name an event, for example "Rebuild Hari Raya 2023 at grandma's." Event Memory Rebuilder then:

1. Finds the matching photos and videos
2. Writes short captions for the photos
3. Transcribes the spoken words in the videos
4. Finds and reads related documents such as invitations or tickets
5. Shows you one complete draft
6. After you approve, creates a photo album and saves a written recap combining captions, quoted passages from the videos, and document details

Original files are never changed, moved or deleted. If something is missing (a silent video, an unreadable document, a photo with no date), it says so and leaves it out instead of guessing.

How I built it

The Skill is written entirely in Markdown following the Uliya Skill Development Standard. It has no scripts and no outside services.

• I used the UGOS Skill Creation Tool and its capability catalog to confirm what the NAS can do before writing anything.
• Each step passes only real results to the next one.
• Claude helped me read the kit, draft the Skill and review it.
• I checked the package with the official static checker.

Challenges I ran into

• Designing around real limits. The catalog has no blur detection, no way to describe what happens inside a video, and no subtitle-file creation. I dropped those ideas and narrowed video support to spoken words only.
• An open question about video. The transcription example in the catalog uses an audio file. Whether video files are accepted is unconfirmed, so the Skill reports unreadable videos and keeps going instead of failing.
• Strict format rules. The checker is picky about section names and how capabilities are listed, so I had to read its rules closely to satisfy them.
• Gaps found in testing. A paper dry run with seven scenarios exposed five real problems, such as document dates falling outside the event window and what to do when the album is created but the recap can't be saved. I fixed all of them.

Accomplishments that I'm proud of

• The package passes the official static checker with 0 errors and 0 warnings.
• It never invents content, and it asks for approval before creating anything.
• It handles empty results, silent videos, undated photos and half-finished runs clearly.

What I learned

• Good Skills start from what the platform can really do, not from the best-sounding idea.
• Asking for approval makes a tool feel safe with family memories.
• Handling empty results and failures matters as much as the happy path.

What's next for Event Memory Rebuilder

• Test on a real NAS: first whether videos can be transcribed directly, then album creation with the exact approved items.
• Raise the five-video limit and support more spoken languages.
• If the platform adds video understanding or subtitle files, add visual scene descriptions and playable subtitles.

Built With

• ai-agents
• claude
• markdown
• nas
• photo-management
• speech-to-text
• ugos
• ugreen-ai-nas
• uliya-skill-kit