Inspiration
The person who keeps the family photo library means every month to show the children's photos to a grandmother who does not use the app. Choosing twelve from three hundred, writing something true about each, and remembering which ones she already has takes an evening nobody spends. Phone highlight features are owner-facing and cycle the same favourites; cloud photo AI wants the whole archive uploaded first.
What it does
Family Memory Letter is a UGOS Skill for the Bring Memories to Life track. Ask it for one recipient and one month, and it returns three linked things: an album holding exactly the photos the letter mentions, a 250-400 word letter whose captions trace back to stored capture time, location and a visual description, and a tag ledger on those photos so next month's letter cannot repeat them.
The unit of work is one recipient and one month, and the artifact is a letter rather than a feed.
How I built it
Orchestration only: one SKILL.md and five reference files, no scripts. Reads run first — the tag ledger, then the person album and a date-bounded photo search, then capture-time metadata and a visual description for each shortlisted photo. Four writes follow, each behind a platform approval: the album, the letter draft, filing it into the family's folder, and the ledger tag. The ledger is written last on purpose, so a run that fails at any step records nothing as sent. Photos stay on the NAS; the local AI does the vision and the writing, and drops to date and presence when it is unsure.
Challenges
Recognition needs curated person albums, and group shots return few candidates. Vision can be wrong, so the letter is editable and uncertainty is dropped rather than smoothed over. A wrong camera clock can land a photo in the wrong month, so those are excluded and reported instead of guessed.
What's next
On the assigned NAS I would test a 300-photo month against an empty ledger: twelve moments, at most three per day, one album, one letter, twelve bindings. Then rerun the same month and expect only unseen photos.
Built With
- ai
- photo
- productivity
