Ofrenda：私人記憶花園

Ofrenda: A Private Memory Garden | Hacktoberfest Week 4

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine.

What I Built

Ofrenda is a private memory garden for keeping the small details of someone’s life close: a story, the dish they made, a song from a familiar drive, or a phrase you can still hear in their voice. You can create a memorial, add a photo and keepsakes, and write a remembrance yourself.

I interpreted “Ghost in the Machine” as a question about what remains with us. Not a digital ghost pretending to be a person, but the way a voice or habit can stay present through the memories people carry. The project is inspired by the season of Día de los Muertos and the idea of gathering meaningful reminders. It is a personal digital space, not a replacement for the traditions, altars, or time together that give those remembrances their meaning.

An optional local AI helper can turn details you choose to save into a first draft. It is a writing aid—not a simulation of the person, and not the authority on their life.

Demo

Live demo: https://tritax.netlify.app

Promo

https://jumpshare.com/s/I6UzBYRavxuvWFXuPuAr

Code

The public source repository is on GitHub.

How I Built It

Ofrenda is built with React, TypeScript, and Vite. Memorial profiles, photos, and saved remembrances live in browser IndexedDB. There are no accounts or profile server. The app also supports JSON export and import so a person can keep a separate backup or move their garden to another browser.

The optional writing feature uses Hugging Face Transformers.js and the open-weight SmolLM2-360M-Instruct model, with its 4-bit ONNX weights. Inference runs in a browser Web Worker. The model is not downloaded until someone chooses Prepare local writing. Its selected weight file is about 388 MB, so the first download needs an internet connection. After the model assets are cached, generation runs in the browser; the memorial text is not sent to a hosted inference API.

The result remains an editable draft. I prompt the model to use only the details supplied and not invent biographical facts, but small models can still be wrong. The person remembering decides what to keep. The garden can also be used without downloading or running the model.

Why Does Open Innovation Matter?

These are intimate stories. A closed AI API would make the writing feature depend on sending those stories to a service outside the person’s control. With open weights and local inference, Ofrenda can offer optional help without making a hosted AI endpoint the keeper—or recipient—of family memories.

Open tools also make the implementation easier to inspect and adapt. The trade-off is real: the model download is large for a small app, browser storage can be cleared, and the model’s English writing can be imperfect. That is why the feature is opt-in, the non-AI experience works on its own, and every generated remembrance must be reviewed by a person.

Prize Categories

I’m entering the overall challenge. Ofrenda does not meaningfully use any of the listed partner technologies, so I’m not claiming a partner category.

DEV tags: devchallenge, hf26challenge