Memory Lantern：以 AI 整理並保存家族故事的記憶花園

• A little light for the stories we never want to lose.

What I Built

What if AI could help us remember without pretending to be the people we miss?

• Memory Lantern is a memory garden built into the Dr. T platform. It helps people preserve family stories, recipes, traditions, meaningful sayings, and moments of everyday joy, then weave selected memories into editable stories and tributes.

For Hacktoberfest Week 4’s “Ghost in the Machine” theme, I chose a gentler interpretation of the ghost: not a digital imitation of someone who has passed away, but the traces people leave behind in their stories, voices, customs, and acts of love.

Each memory becomes a lantern—a small light that can be revisited, organized, and shared intentionally.

The experience includes a memory garden, chronological timeline, storybook, editable AI-assisted narratives, printable remembrance pages, and archive export tools. Fictional sample memories make it possible to explore the experience without exposing anyone’s private family history.

My guiding principle is simple: preserve the memory, not impersonate the person.

Demo

• Live preview: https://ais-dev-4s4jvpipr3mh3mz6x2hpfp-393352619239.asia-southeast1.run.app

Video:

Open the Memory Lantern tab in Dr. T.

The current environment has reported that the Gemma inference provider is not configured. The interface and integration are implemented, but live Gemma-generated stories should not be considered verified until the provider is configured and a real generation succeeds.

• Demo recording: Add a short screen recording here after verifying the complete workflow, including real model generation if configured.

Code

**Platform repository:
https://github.com/Zenieverse/Dr.-T

Memory Lantern was developed as a new module within the existing Dr. T platform, with a standalone memory-lantern/ package prepared for separation.

• Dedicated Memory Lantern repository: Add the public repository URL here once it has been created and verified.

The implementation includes the memory domain types, UI components, memory service, server-side Gemma provider adapter, privacy documentation, and automated verification tests. The standalone package and its licensing and attribution should be checked before publication.

How I Built It

Memory Lantern was built with TypeScript, React, the existing Dr. T application architecture, and a server-side AI provider adapter.

The open-weight AI integration targets Gemma 2, with support designed for local Ollama inference or a compatible hosted, OpenAI-style inference endpoint.

The intended workflow is straightforward:

1. A user selects memories they want to preserve.
2. The application sends the selected material to the server-side story-generation endpoint.
3. The configured Gemma model generates an editable narrative.
4. The user reviews and edits the result before saving it as a new lantern.

The server-side prompt includes constraints against impersonating deceased people and inventing biographical facts. The interface is designed to distinguish user-supplied details from narrative enhancements.

The implementation also includes memory editing, search, timeline and storybook views, print-friendly reading, and export functionality. Existing Dr. T navigation and modules were preserved.

The supplied development audit reports successful TypeScript checking, a production build, and 281 passing assertions across ten test suites. Those figures still need to be reconciled with the individual suite totals before being presented as final verified results.

Important current limitation: the hosted preview reported NOT_CONFIGURED for Gemma inference during the audit. Therefore, the integration is implemented, but successful live inference has not yet been established. I will update this section when a real Gemma generation has been verified.

Why Does Open Innovation Matter?

Personal stories are too important to treat AI as a mysterious black box.

Open-weight models offer developers more choices about where inference runs, how a system is configured, and how its behavior can be evaluated. A local Gemma deployment can offer a different privacy trade-off from sending memories to a hosted provider, although local inference still requires appropriate device security and data handling.

Open innovation also makes it easier for others to inspect the architecture, reproduce the project, improve the safety constraints, and adapt the experience to different families, languages, and cultural traditions.

For Memory Lantern, the point is not to generate a convincing imitation of a person. It is to help people organize what they genuinely remember, while keeping the human in control of the story.

The Spirit of It

A grandmother’s recipe is more than a list of ingredients. It might be the smell of tea in the kitchen, the way someone measured without a spoon, or the words they always said while serving it.

Memory Lantern is for preserving those details without pretending that a machine can bring the person back.

The AI can help arrange the words. The memories, meaning, and final choices belong to the people who lived them.

• A lantern does not replace the light that came before it. It helps us find our way back to it. 🏮

Categories

Add only the partner prize categories for which the project meets the published eligibility requirements. Do not claim a category based solely on the presence of an integration; verify the rules and demonstrate the required functionality first.

Acknowledgments

Memory Lantern is a new project developed within the Dr. T platform. Existing platform infrastructure and reused components should be credited clearly, alongside the licenses of any third-party libraries and model assets.