Recipe Keeper - Hacktoberfest '26

Ashish Anand — Mon, 05 Oct 2026 03:32:50 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

Recipe Keeper is a small web app that turns [Grandpa's / Name's] rambling voice memos into a printable family cookbook.

I built it for my grandfather. He cooks entirely by feel: "a fistful of rice, no, more, and onions until it smells right." None of it is written down, and the recipes will be lost if nobody captures them. A normal recipe app would force exact measurements he never uses, and it would throw away the stories that make the food his.

You paste in a transcript of what he said. A local AI model turns it into a clean recipe card with a title, ingredients, steps and notes. It follows two rules:

It never invents quantities. "A fistful" stays "a fistful".
Unclear parts become questions to ask him, such as "Ask Grandpa: how long is 'until it smells right'?"

Every card also keeps a quote of his original words underneath. Each recipe has a photo slot for the dish, and the whole cookbook can be printed or saved as a PDF to give back to the family.

Demo

https://youtu.be/6sBEymtBEBc

Code

https://github.com/ASAN-11/Recipe-Keeper

It's a single file, recipe-keeper.html, with no build step and no dependencies.

How I Built It

Model: an open-weight model run locally with Ollama (I used llama3.2, and the app lets you pick any model you have installed).
Inference: the page calls Ollama's local /api/chat endpoint. Responses are streamed so you see the model working, and JSON mode (format: "json") with a low temperature keeps the output a structured recipe.
Prompting: the system prompt tells the model to use only what was said and never guess amounts. It also tells it to flag unclear steps as questions for the cook.
App: plain HTML, CSS and JavaScript. The cookbook is saved in the browser's local storage, and it can be exported or imported as a JSON backup. The hero section takes a background photo or video, and a print stylesheet makes a clean PDF.
Setup: ollama pull llama3.2, then serve the page from localhost (python -m http.server) or start Ollama with OLLAMA_ORIGINS="*".

Why Does Open Innovation Matter?

Privacy: these are family recipes and a grandparent's own voice and stories. With a local model, none of it goes to a server I don't control, and there's no account to make.
Works offline: after the one-time model download, it runs with no internet. That matters in a kitchen or at a grandparent's house.
Free to run: there are no API keys and no per-token cost, so I can process as many memos as I want.
Swappable and tunable: I can change models from a dropdown, and later I could fine-tune one on the finished cookbook so the cards sound like him.
I control the behavior: the "never invent quantities" rule matters a lot for heirloom recipes. With an open model I own the prompt and settings, and no provider can change them under me.