Yaadein: a private remembrance maker for a loved one, powered by open-source Gemma

I built this for the spirit of Hacktoberfest 2026's final theme, Ghost in the Machine, but I finished it after the deadline, so it is not a challenge entry.

What I Built

Yaadein ("Memories") is a small web app for remembering someone you love, inspired by the ofrenda in Día de los Muertos. You enter their name, favourite foods and songs, and a short memory. A local AI model writes a warm remembrance in English, Hindi or Hinglish, shown on a candlelit card with the foods and songs as little chips. You can edit the text, copy it, or print it or save it as a PDF.

The ghost in this machine is memory: the model never pretends to be the person. It only helps you find words for your own memory of them.

Demo

• Live page (interface only): https://saurabhh-cloud.github.io/yaadein/
• The AI runs locally, so the page needs Ollama on your own computer to write the text. Setup steps are in the README. ## Code

saurabhh-cloud / yaadein

Yaadein 🕯️

A quiet place to remember someone you love. Enter their name, favourite foods and songs, and a short memory, and a local AI model writes a warm remembrance you can edit, print or save as PDF, shown on a candlelit "ofrenda" card.

Made in the spirit of Día de los Muertos and the Hacktoberfest 2026 final theme, "Ghost in the Machine". Built after that challenge closed, so it is not a challenge entry.

Open-source AI at the core: Google's open-weight Gemma model runs on your own computer through Ollama. Nothing about your loved one is uploaded anywhere, and it works with the internet off.

Run it

1. Install Ollama: https://ollama.com/download
2. ollama pull gemma3:1b
3. Start Ollama with the page allowed
   • Windows PowerShell: Stop-Process -Name "ollama*" -Force; $env:OLLAMA_ORIGINS="*"; ollama serve
   • Mac/Linux: OLLAMA_ORIGINS="*" ollama serve
4. Open index.html in Chrome (or the GitHub Pages link) and click Light a candle.

The…

View on GitHub

How I Built It

• Frontend: one HTML file with plain CSS and JavaScript.
• AI: Google's open-weight Gemma (gemma3:1b) running locally through Ollama, with the reply streamed word by word.
• Prompt: the model is told to use only the facts I typed, not to invent dates, places or relatives, not to pretend to be the person, and to output only the remembrance.
• Editing: the text becomes editable after it is written, because a remembrance should be corrected by the person who knew them.

• Printing: a print style hides the form so the card prints cleanly or saves as a PDF.

  Why Does Open Innovation Matter?

• Privacy: these are the most personal stories a family has. With a local model they never leave the computer.

• It works with the lights off: no internet, no account, no subscription that could disappear.

• Control: I can change the model or the tone of the prompt myself.
  Where it falls short: gemma3:1b is small, so the writing can be plain, and its Hindi may not be perfect. I wrote and tested the app with a stand-in server for the model, and have not yet run it end to end with real Gemma, so the quality of the real remembrances is untested. Setup also needs Ollama, which is a barrier for non-technical families.