***Delivering the weather in a more human way with [Ben](https://www.bencobley.com)*** ![[inkyweather.gif]] When I check the weather forecast, it's really because I'm wondering **what to wear** and if the weather will **affect my plans**. But, when you open the weather app you see a collection of symbols and numbers: 9AM: 10˚ Rain | 12PM:11˚ Sunny | 3PM: 13˚ Light Rain | 6PM: 12˚ Storm While they're still useful, wouldn't it be better to get to the point? [Ben](https://www.bencobley.com) and I developed Inky Weather to create this forecast in a memorable sentence and display it on e-ink, so you can catch a glimpse before you leave in the morning - you’ll know whether to layer up, layer down or grab an umbrella. ### Building Inky Weather We started this project before chatGPT and LLMs took off by hand-crafting the sentence structures. Soon enough, we discovered OpenAI's GPT3 and realised GPTs do a better job (when they aren't hallucinating!). Through many iterations and lots of data wrangling, we engineered reliable prompts that give accurate results. Inky Weather is now running on GPT4. ![[inkygif.gif]] ## So, how does it work? 1. Get and format the weather forecast via the Met Office API. 2. Prompt GPT4 via the OpenAI API to create a weather forecast from the data. 3. Style the forecast as needed with a secondary prompt. 4. Display on an inkyWHAT display with Raspberry Pi. ![[IMG_6200.jpg ]] >[!tip] >You can also style the weather according to a character or persona! > ![[IMG_5429.png]] --- [![[Pasted image 20240110213631.png | 100]]](<https://github.com/mimireyburn/Inky-Weather>)