What I Built
I built Ofrenda, a private, offline-first memory keeper for preserving the small details that make someone unforgettable.
You can add stories, favorite foods, songs, rituals, and meaningful dates connected to a loved one. Ofrenda turns those memories into a personal remembrance card and a digital Día de los Muertos ofrenda.
The “ghost” in my project is the feeling of someone still being present in their recipes, songs, routines, and stories.
Demo Video
Watch the project demo:
https://www.youtube.com/watch?v=o8I3ROGjlfI
Try the live app:
https://ofrenda-3cwxygxg.manus.space
How I Built It
Ofrenda stores memories locally in the browser. It does not require an account, database, or cloud API.
The remembrance composer uses a replaceable local-model adapter:
LocalOfrendaAdapter.generate({
memory,
tone,
include
})
The same adapter can connect to WebLLM, llama.cpp, a browser-cached open-weight model, or a local inference server.
The current demo includes an offline-safe fallback so the experience works even when model weights are not downloaded.
Why Open Innovation Matters
The memories we most want help with are often the memories we least want to upload.
A family story, recipe, photograph, or personal detail should not automatically disappear into a remote API.
Open innovation makes it possible to:
- Keep memories on the user’s device
- Work without an internet connection
- Swap models instead of being locked to one provider
- Choose models that understand different languages and cultures
- Keep the project useful even if a hosted service changes or disappears
For Ofrenda, local-first AI is not just a technical choice. It is part of the emotional design.
The machine can help shape the story.
But the memory remains yours.
The Spirit of It
Someone opens Ofrenda and sees a familiar name, a favorite song, a recipe, and a ritual they had almost forgotten.
Then the machine helps turn those fragments into a few sentences.
Not to replace the person.
To help keep the person close.
Top comments (0)