Grandma’s Kitchen Notebook

#devchallenge #weekendchallenge #hf26challenge

Hacktoberfest Weekend Challenge: Build for a Friend Submission 🤝

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

What I Built

I built Grandma’s Kitchen Notebook, an AI-powered recipe memory app for my grandmother.

She has many voice notes and handwritten recipe cards with family dishes, tips, and stories. The problem is that these memories are hard to organize and easy to lose. I wanted a simple app where she (and our family) can upload audio and recipe images, then get clean, structured recipes.

The app:

accepts voice notes (.mp3/.wav/.m4a)
accepts recipe card images (.png/.jpg/.jpeg)
transcribes and extracts text
generates structured recipe output (title, ingredients, steps, story)
stores entries in a local SQLite cookbook archive

Demo

Live app:

👉 https://grandmas-kitchen-notebook-kxyqyvchgryaohlre5abt9.streamlit.app/

Code

GitHub repository:

👉 https://github.com/AslamSujah/grandmas-kitchen-notebook

How I Built It

This project is built around open-source AI components:

faster-whisper for speech-to-text transcription
Tesseract OCR (pytesseract) for extracting text from handwritten recipe images
Gemma (via Ollama) for recipe structuring when an LLM endpoint is configured
Cloud-safe fallback parser so the app still works even when Ollama is not available
Streamlit for the user interface
SQLite for local recipe storage

Pipeline flow:

Upload audio/image
Audio transcription + image OCR
Merge extracted text
Parse into structured recipe JSON
Save and display in recipe history

Why Does Open Innovation Matter?

Open innovation made this project practical and meaningful:

Model choice freedom: I can use open-weight models like Gemma and switch models as needed.
No vendor lock-in: The pipeline is modular and can run with different OSS tools.
Cost control: No mandatory per-call closed API billing.
Privacy-first path: Same architecture can run in local/offline mode for sensitive family data.
Customizability: Prompting/parsing behavior can be adapted for dialect-heavy or family-specific cooking language.

A closed API-only approach would be less flexible, harder to customize, and more expensive for repeated family-archive processing.