Fridge Forecast: a weekly outdoor plan you print once and never look up again

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

Apps that tell you to go outside have an obvious flaw: you have to open an app.

Fridge Forecast moves the whole thing onto paper. Once a week, you export your calendar and run one command. It finds the free windows in your week, checks the forecast for each, and prints a one-page card for the fridge showing the best times to go outside and what to do with them. After that, the screen's job is finished.

Demo

The flow has three steps:

1. Export your calendar as an .ics file and add a line about what you like.
2. Run Fridge Forecast. It picks the best windows and writes a suggestion for each.
3. Print the card and put it on the fridge.

How I Built It

Everything runs locally except the forecast lookup:

• Calendar: the icalendar library reads your .ics file and works out free windows.
• Forecast: Open-Meteo, a free open API that needs no key.
• Choosing windows: plain code scores each window on rain, wind, temperature, and daylight. The model never decides when you go out.
• Suggestions: an open-weight Gemma model served through Ollama writes one friendly suggestion per chosen window, using your note about what you like.
• Output: a printable HTML page.

The key design choice is the split. Code makes the decisions that must be right, such as when the weather is good and when you're free. The model only writes the human part. Its output is validated, and if it fails, fixed fallback text keeps the card complete.

Why Open Source AI

• Privacy: your calendar shows when you're home, where you are, and who you meet. With a local model, it never leaves your machine.
• Cost: it costs nothing to run, so it can be a weekly habit with no subscription.
• Control: the scoring rules and the prompt are plain files you can edit to suit your own life.

Comments

[Rajdeep Dolui]

Excited to be building for this one! 🌿

My entry is Sit Spot, a voice-note nature journal where the screen is the shortest part. You sit outside for ten minutes and talk through what you see and hear, and back inside it turns the recording into a Markdown journal entry in your own words.

It's all open source and runs locally: Whisper for transcription, Gemma through Ollama for the writing, and Open-Meteo for the weather. Your voice and notes never leave your machine.

Repo: github.com/bidyutdolui1970-cpu/sit...

Looking forward to seeing everyone else's takes on getting people outside! 🧭