DAS IST BLITZKRIEG
Four finished coding projects and a 39-page research paper. September was a lot.
I had just released Windows Never Forgets at the end of August. A 162-page book was already something I was proud of, but I still had Cicero sitting there, abandoned twice, and eight months of studying AI that I wanted to put to use.
On September 12th, Cicero was finally finished! Its third version became an agentic framework for translating long documents. I spent two days fixing the orchestration after getting the skeleton working. Apparently, getting agents to do their jobs in the right order was going to take more patience than I expected.
The next day, I finished adLibere. All I wanted was to watch Downfall, and somehow I ended up coding an adblocker over the weekend. The gambling ads also left me with a question that kept bothering me after the extension was done: how accessible had all of this become?
I started collecting material for that research around the 15th. Meanwhile, the question bank my tutor, Miragha Gafarzada, had provided gave me what I needed to work on the SAT Math generator we'd discussed back in May. During coding, I realized that the same approach could work with question datasets for other exams. By September 22nd, it was Gaussian PREP Framework.
Then came Ghostnector, and I gave myself a rather interesting problem: I wanted to build it in Rust, a language I could read but couldn't code in.
For the first time, I had AI write the whole project while I planned the modules, supervised the implementation, and debugged. Ten modules, multiple checkpoints, four days of intensive work. Getting to v1.0.0 felt good, especially after all the testing. I also went without sleep for two nights, which is a part of the process I'd rather not repeat.
And that question from the movie site? By the 29th, it had become a 39-page research paper. I went through data on gambling participation, advertising, financial vulnerability, and much more. One result that stayed with me was how strongly marketing and gambling revenue appeared connected until I looked at their month-to-month changes. It gave me plenty to think about before writing conclusions.
Looking at the dates now is funny. Cicero had already failed twice. Gaussian had waited for months. adLibere wasn't even planned. Then all four projects were finished within the same month, with the research happening alongside them.
I'm proud of this September. There were enough abandoned attempts and hours of debugging behind it that seeing the finished work together feels a little strange.
Now I have a circadian rythm to fix. Thanks for reading!!