I finished the Day 1 challenge. But I didn't want to just connect the tools and stop there. I spent more time on the model choices and on the output quality, because I want to really use this.
The biggest thing I learned: first I used the image model to create the infographics. This was a mistake. Image models are not good with numbers. The chart looks nice, but the numbers on it are wrong. For a newsletter with sources, this is really bad.
So I changed it. Now the charts come from the real research data, and the AI image model only makes the header picture. Correct numbers, and still nice design.
I also picked cheaper models for each API instead of the biggest one. Cost is about 4 cents per newsletter.
Then I made a version for my target market, Turkiye. 🇹🇷
This part was interesting. My first Turkish version was correct, but it was boring to read. It felt like English text in Turkish clothes.
The problem was not the words. The words were correct, but the meaning was wrong in context. So I used Claude to write it again in Turkish from zero. Not translated, written. And I saved these rules in my workflow, so it will not happen again.
Screenshots: the English one, my first Turkish try, and the new version. Look at the difference between 2 and 3.
My takeaway: the automation part was easy. The difficult part was learning what the tools cannot do, and that "translated" and "written" are not the same thing.