MasakApa: Building a Local AI Meal Planner for My Wife

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

What I Built

I built MasakApa for my wife, who works and also cares for our children.

After a long day, deciding what to cook can become another source of stress. MasakApa helps by turning the ingredients already available at home into practical meal ideas.

The user enters available ingredients, selects a cooking mood, and chooses the number of servings. MasakApa then generates three recipe suggestions, including:

• Why the recipe fits the situation
• Available ingredients
• Required missing ingredients and seasonings
• Optional additions
• Simple cooking instructions
• Copy and download actions

Demo

Watch the MasakApa demo recording

Code

View the MasakApa source code on GitHub

How I Built It

MasakApa is built with Python and Streamlit. It uses Gemma 3 4B through a local Ollama server and the OpenAI-compatible API.

The application flow is:

Streamlit UI
    ↓
Application state and meal services
    ↓
OpenAI-compatible client
    ↓
Ollama → Gemma 3 4B

The app includes defensive response parsing because local models may return Markdown, explanations, thinking blocks, or imperfect JSON. It also validates recipe data and ensures that missing seasonings are included in the cooking instructions.

Why Does Open Innovation Matter?

Using a local open-weight model made it possible to build a private and accessible household tool.
Ingredient information can remain on the user’s computer instead of being sent to a hosted AI service. The model, prompt, and response-processing logic are also visible and changeable.
This makes the solution easier to understand, adapt, and run without depending on a paid closed API.

Prize Categories

• Best Use of Gemma — Gemma 3 4B running locally through Ollama