Inspiration
In Nigeria, every big family occasion (a 70th birthday thanksgiving, a retirement, a burial) comes with a printed programme brochure that every guest takes home. Someone always ends up building it in one stressful week: photos scattered across phones and old prints, dates nobody is sure of, voice notes buried in WhatsApp, and songs on a flash drive. Facts get guessed and printed wrong. Everything needed is usually already on the family's devices.
What it does
Ìrántí ("remembrance" in Yoruba) is a UGREEN AI NAS Skill. A request, such as "Prepare Ìrántí for Daddy's 70th", produces:
- a photo album covering every decade of the person's life, with honest captions
- a life story where every fact cites the document it came from
- quotes in the person's own words from their voice notes, used only with family consent
- a programme brochure draft, with an optional Yoruba or Igbo edition
- a list of gaps relatives can still fill ("no photos from the 1970s")
- a run-of-show: on the day, "start the cake cutting" plays the right song from the family's own music library
There are two modes: Celebrate for birthdays, retirements and anniversaries, and Remember for funerals and memorials. Remember mode adds a yearly album that resurfaces on the date of passing.
How I built it
The skill is pure markdown, following the Uliya skill standard. It orchestrates 18 delegate calls across five NAS experts (photo, file, document, knowledge and media). Everything is read-only until the family approves a write, and four checkpoints keep a person in charge of the sensitive decisions: confirming the face, choosing the voice notes, approving quotes, and settling conflicting dates.
Challenges
- Old family photos are scanned, so their recorded date is the scan date, not 1984. Ìrántí dates prints from writing on the photo, then from the family, and otherwise says "date unknown".
- Making sure the AI never "improves" a life story with invented facts.
- Handling voice notes respectfully: consent first, and no transcripts are saved.
What's next
Testing on a real UGREEN NAS with a relative's consented photos and voice notes, and improving Yoruba, Igbo and Pidgin transcription.
Built With
- claude-code
- document-agent
- file-agent
- knowledge-agent
- markdown
- media-agent
- ocr
- photo-agent
- speech-to-text
- ugos
- ugreen-ai-nas
- uliya-skill-kit
- vision-language-model
