Hearth (LegacyLink)

I built Hearth because I know what it feels like to realize there are stories I can no longer ask my grandparents to tell. They passed away before I had captured some of the experiences only they could describe. Later, parts of those stories reached me through other people—but a secondhand retelling can lose details or change with time. Hearing a story from the person who lived it is not the same as hearing someone else's version.

Hearth is my attempt to make preserving those memories feel less like a task and more like a conversation. A family member can speak naturally; Hearth listens, asks gentle follow-up questions, and saves the memory so relatives can return to it later. Families can retrieve stories, play quizzes based on what was shared, and leave messages for future generations. The goal is not to replace family conversation. It is to help important memories survive it.

The local prototype combines a browser voice experience, a Node.js conversation service, and an MCP server with ten oral-history tools. It supports seven languages and includes a Python runtime intended for a standalone Raspberry Pi device. The core web and conversation flows work locally. I have not yet packaged a finished appliance or published a live deployment.

The current automated JavaScript run passed 9 of 9 tests. The Python device suite ran 7 tests: 5 passed, 1 failed, and 1 was skipped in my latest Windows run; the simulation test did not receive its expected reply, so I am treating device validation as incomplete. The local prototype has been exercised, but the physical Raspberry Pi audio/peripheral path is not yet certified.

I also implemented optional integration paths for Amazon Bedrock, Polly, and DynamoDB. I have not validated those services against a live AWS account. My AWS account setup is blocked because the phone-verification code has not arrived despite repeated attempts. I have contacted AWS Support and am waiting for help. The app's local mode does not require AWS.

I learned how to structure oral-history conversations as reusable MCP tools, build multilingual flows, and test the same conversation path across the web app and device runtime. The next steps are to publish and deploy a judge-ready web demo, validate AWS integrations if account verification is resolved, test the runtime on actual Raspberry Pi hardware, and improve privacy protections before real families rely on it.