CICERO IS FINISHED
Cicero was on the table not because it suddenly came to my mind, but because it was already the project that I had failed to finish twice.
I started studying AI and how it works in January of 2026. Eight months of preparation gave me enough confidence to code my very first project.
Cicero was on the table not because it suddenly came to my mind, but because it was already the project that I had failed to finish twice.
In its first version, Cicero was going to be a lightweight live web translator for social media platforms. How silly of me, of course such a project was not going to be a lightweight Chrome extension! I abandoned it within the first ten days of July.
Later, I decided to code Cicero as a locally trained translation model, but abandoned it quickly: I lacked the computing power needed to undertake such a project, plus I still wasn't sure that I was ready for model training.
And finally, around September 8th, I decided to code it as an agentic framework to translate long documents: a common issue for chat models because of their context-window limitations.
Spending 5–6 hours a day, I managed to finish the skeleton of the code around the 10th. What did I do for the remaining two days?? Tried to fix the orchestration of the agents!
It was a funny mistake to make on my first try, as it took me two days to fix completely and confirm the security audit, but it was totally worth it!
Congrats to myself!!